{ "schema_version": "1.0.0", "build": "development", "rows": [ { "id": "cuda-vram-rtx-4090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-4090/wan22-t2v-a14b/vram//", "axis": "vram", "model": "wan22-t2v-a14b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 22.188, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-vram-rtx-5090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-5090/wan22-t2v-a14b/vram//", "axis": "vram", "model": "wan22-t2v-a14b", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 22.19, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-steptime-rtx-4090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-4090/wan22-t2v-a14b/step_time//", "axis": "step_time", "model": "wan22-t2v-a14b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 32.382, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-steptime-rtx-5090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-5090/wan22-t2v-a14b/step_time//", "axis": "step_time", "model": "wan22-t2v-a14b", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 26.093, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-render-rtx-4090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-4090/wan22-t2v-a14b/render_time//", "axis": "render_time", "model": "wan22-t2v-a14b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 1303.655, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-render-rtx-5090-wan22-t2v-a14b-ultrafast", "cell": "nvidia-cuda/rtx-5090/wan22-t2v-a14b/render_time//", "axis": "render_time", "model": "wan22-t2v-a14b", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 1050.018, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-20", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "capybara-wan-2026-10", "condition": "Dedicated single-GPU cloud host; one clean render; board power as recorded in the evidence file.", "dev_build": "WAN 2.2 A14B development build (2026-10-20)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen35-0.8b-512", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen35-0.8b/decode/512/", "axis": "decode", "model": "qwen35-0.8b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 512, "out": null, "value": 688.1, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA A100-SXM4-80GB; board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen35-2b-512", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen35-2b/decode/512/", "axis": "decode", "model": "qwen35-2b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 512, "out": null, "value": 439.4, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA A100-SXM4-80GB; board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen35-4b-512", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen35-4b/decode/512/", "axis": "decode", "model": "qwen35-4b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 512, "out": null, "value": 224.3, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA A100-SXM4-80GB; board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen35-0.8b-512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen35-0.8b/decode/512/", "axis": "decode", "model": "qwen35-0.8b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 512, "out": null, "value": 1003.6, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen35-2b-512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen35-2b/decode/512/", "axis": "decode", "model": "qwen35-2b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 512, "out": null, "value": 660.2, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen35-4b-512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen35-4b/decode/512/", "axis": "decode", "model": "qwen35-4b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 512, "out": null, "value": 365.1, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen35-9b-512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen35-9b/decode/512/", "axis": "decode", "model": "qwen35-9b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 512, "out": null, "value": 241.6, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); board power not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen35-9b-512", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen35-9b/decode/512/", "axis": "decode", "model": "qwen35-9b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 512, "out": null, "value": 129.45, "unit": "tok/s", "runs": 2, "min": 127.0, "max": 131.9, "spread_pct": 3.79, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA A100-SXM4-80GB; board power not recorded; dedicated single-GPU cloud host; no quiet gate; two runs on separate pods; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen36-35b-a3b-512", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/decode/512/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 512, "out": 64, "value": 272.53, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token context, 64 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 4090; board power not recorded; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/512/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 512, "out": 64, "value": 268.82, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); board power not recorded; driver 580.126.09; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen36-35b-a3b-512", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen36-35b-a3b/decode/512/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 512, "out": 64, "value": 192.05, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token context, 64 generated tokens, batch 1", "condition": "NVIDIA A100 80GB; board power not recorded; driver not recorded; dedicated single-GPU cloud host; no quiet gate; one run; a measurement of the shipped package on this card will replace it.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-step-rtx-3090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3477, "unit": "s/step", "runs": 3, "min": 0.34756, "max": 0.34877, "spread_pct": 0.35, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.41, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3461, "unit": "s/step", "runs": 3, "min": 0.34557, "max": 0.34608, "spread_pct": 0.15, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.08, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2761, "unit": "s/step", "runs": 3, "min": 0.27589, "max": 0.27616, "spread_pct": 0.1, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.35, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.276, "unit": "s/step", "runs": 3, "min": 0.27575, "max": 0.27627, "spread_pct": 0.19, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.35, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 1.5117, "unit": "s/step", "runs": 3, "min": 1.51151, "max": 1.51264, "spread_pct": 0.07, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.8314, "unit": "s/step", "runs": 3, "min": 0.83128, "max": 0.83156, "spread_pct": 0.03, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.8279, "unit": "s/step", "runs": 3, "min": 0.82781, "max": 0.82846, "spread_pct": 0.08, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.66, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.5584, "unit": "s/step", "runs": 3, "min": 0.55838, "max": 0.55877, "spread_pct": 0.07, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-3090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.5588, "unit": "s/step", "runs": 3, "min": 0.55865, "max": 0.55884, "spread_pct": 0.03, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-3090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1687, "unit": "s/step", "runs": 3, "min": 0.16821, "max": 0.16872, "spread_pct": 0.3, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.68, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1675, "unit": "s/step", "runs": 3, "min": 0.16681, "max": 0.16772, "spread_pct": 0.54, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.74, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1297, "unit": "s/step", "runs": 3, "min": 0.12958, "max": 0.12995, "spread_pct": 0.29, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.62, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1297, "unit": "s/step", "runs": 3, "min": 0.12918, "max": 0.13021, "spread_pct": 0.79, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.62, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.8731, "unit": "s/step", "runs": 3, "min": 0.87096, "max": 0.87346, "spread_pct": 0.29, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4226, "unit": "s/step", "runs": 3, "min": 0.42234, "max": 0.42356, "spread_pct": 0.29, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4179, "unit": "s/step", "runs": 3, "min": 0.41774, "max": 0.41812, "spread_pct": 0.09, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.8, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2701, "unit": "s/step", "runs": 3, "min": 0.26994, "max": 0.27064, "spread_pct": 0.26, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-4090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2704, "unit": "s/step", "runs": 3, "min": 0.26995, "max": 0.27094, "spread_pct": 0.37, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-4090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1275, "unit": "s/step", "runs": 3, "min": 0.12745, "max": 0.12753, "spread_pct": 0.06, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1275, "unit": "s/step", "runs": 3, "min": 0.12741, "max": 0.12763, "spread_pct": 0.17, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.59, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1104, "unit": "s/step", "runs": 3, "min": 0.11033, "max": 0.11045, "spread_pct": 0.11, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.86, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1105, "unit": "s/step", "runs": 3, "min": 0.11039, "max": 0.11048, "spread_pct": 0.08, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.86, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3123, "unit": "s/step", "runs": 3, "min": 0.31233, "max": 0.31234, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.03, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3122, "unit": "s/step", "runs": 3, "min": 0.3122, "max": 0.31228, "spread_pct": 0.03, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.92, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2308, "unit": "s/step", "runs": 3, "min": 0.23067, "max": 0.23173, "spread_pct": 0.46, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.03, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-5090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2308, "unit": "s/step", "runs": 3, "min": 0.23068, "max": 0.23101, "spread_pct": 0.14, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-5090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.03, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a100-80gb-sxm4-sd3.5-medium-quality", "cell": "nvidia-cuda/a100-80gb-sxm4/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": null, "out": null, "value": 62.76, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100-SXM4-80GB (81920 MiB, sm_80); power 400 W, uncapped; driver 580.159.03; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-sxm4-sd3.5-medium-balanced", "cell": "nvidia-cuda/a100-80gb-sxm4/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": null, "out": null, "value": 31.32, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100-SXM4-80GB (81920 MiB, sm_80); power 400 W, uncapped; driver 580.159.03; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-sxm4-sd3.5-medium-fast", "cell": "nvidia-cuda/a100-80gb-sxm4/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": null, "out": null, "value": 30.73, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100-SXM4-80GB (81920 MiB, sm_80); power 400 W, uncapped; driver 580.159.03; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-sxm4-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/a100-80gb-sxm4/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": null, "out": null, "value": 29.47, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100-SXM4-80GB (81920 MiB, sm_80); power 400 W, uncapped; driver 580.159.03; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-step-a100-80gb-pcie-sd3.5-large-fast", "cell": "nvidia-cuda/a100-80gb-pcie/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a100-80gb-pcie", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4887, "unit": "s/step", "runs": 2, "min": 0.48651, "max": 0.49094, "spread_pct": 0.91, "bound": "exact", "build": "development", "measured": "2026-10-16", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100 80GB PCIe (81920 MiB, sm_80); power 300 W, uncapped; driver 595.91.07 / 580.126.20 / 580.159.03 (varies by pod); dedicated single-GPU cloud host; no quiet gate. Each run is a single render on a separate pod (pod-to-pod spread about 2%).", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-step-a100-80gb-pcie-sd3.5-large-ultrafast", "cell": "nvidia-cuda/a100-80gb-pcie/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a100-80gb-pcie", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4965, "unit": "s/step", "runs": 2, "min": 0.4868, "max": 0.50614, "spread_pct": 3.9, "bound": "exact", "build": "development", "measured": "2026-10-16", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100 80GB PCIe (81920 MiB, sm_80); power 300 W, uncapped; driver 595.91.07 / 580.126.20 / 580.159.03 (varies by pod); dedicated single-GPU cloud host; no quiet gate. Each run is a single render on a separate pod (pod-to-pod spread about 2%).", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-pcie-sd3.5-large-balanced", "cell": "nvidia-cuda/a100-80gb-pcie/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a100-80gb-pcie", "target": "nvidia-cuda", "in": null, "out": null, "value": 54.79, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100 80GB PCIe (81920 MiB, sm_80); power 300 W, uncapped; driver 595.91.07; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-pcie-sd3.5-large-fast", "cell": "nvidia-cuda/a100-80gb-pcie/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a100-80gb-pcie", "target": "nvidia-cuda", "in": null, "out": null, "value": 48.6, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100 80GB PCIe (81920 MiB, sm_80); power 300 W, uncapped; driver 595.91.07; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-e2e-a100-80gb-pcie-sd3.5-large-ultrafast", "cell": "nvidia-cuda/a100-80gb-pcie/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a100-80gb-pcie", "target": "nvidia-cuda", "in": null, "out": null, "value": 45.87, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-12", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA A100 80GB PCIe (81920 MiB, sm_80); power 300 W, uncapped; driver 595.91.07; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sealed ship line (Oct 12 and 16)" }, { "id": "cuda-dev-render-rtx-4090-wan21-t2v-1.3b-balanced", "cell": "nvidia-cuda/rtx-4090/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 162.9, "unit": "s", "runs": 2, "min": 162.497, "max": 163.386, "spread_pct": 0.55, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 6, seed 7", "condition": "NVIDIA GeForce RTX 4090; 450 W limit, mean draw 446.95 W; driver 580.126.20; dedicated single-GPU cloud host; two runs.", "dev_build": "WAN 2.1 1.3B development build (2026-10-14)" }, { "id": "cuda-dev-render-a100-80gb-sxm4-wan21-t2v-1.3b-balanced", "cell": "nvidia-cuda/a100-80gb-sxm4/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": null, "out": null, "value": 192.1, "unit": "s", "runs": 2, "min": 191.905, "max": 192.375, "spread_pct": 0.24, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 6, seed 7", "condition": "NVIDIA A100-SXM4-80GB; 400 W limit, mean draw 387.65 W; driver 580.159.04; dedicated single-GPU cloud host; two runs.", "dev_build": "WAN 2.1 1.3B development build (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-3090-qwen36-35b-a3b-8", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/decode/8/512", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 8, "out": 512, "value": 224.37, "unit": "tok/s", "runs": 3, "min": 224.2382, "max": 224.8, "spread_pct": 0.25, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "8-token prompt, 512 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g13 (2026-10-06)" }, { "id": "cuda-dev-decode-rtx-3090-qwen36-35b-a3b-4000", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/decode/4000/129", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 4000, "out": 129, "value": 211.8, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "4,000-token real-text context, 129 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g4 (2026-10-06)" }, { "id": "cuda-dev-prefill-rtx-3090-qwen36-35b-a3b-8000", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/prefill/8000/", "axis": "prefill", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 8000, "out": null, "value": 9064.9, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "8,000-token prompt, warm", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g10 (2026-10-06)" }, { "id": "cuda-dev-prompt-rtx-3090-qwen36-35b-a3b-8000", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/prompt_latency/8000/1", "axis": "prompt_latency", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 8000, "out": 1, "value": 949.5, "unit": "ms", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "8,000-token prompt, warm, to the first generated token", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g10 (2026-10-06)" }, { "id": "cuda-dev-resident-rtx-3090-qwen36-35b-a3b", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/resident_memory//", "axis": "resident_memory", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 21.85, "unit": "GiB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "after the weights are loaded", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-throughput-rtx-3090-qwen36-35b-a3b-c4", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/throughput/64/", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 64, "out": null, "value": 372.8, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64-token prompts", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-throughput-rtx-3090-qwen36-35b-a3b-c8", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/throughput/64/", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 64, "out": null, "value": 408.2, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64-token prompts", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-throughput-rtx-3090-qwen36-35b-a3b-c24", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/throughput/64/", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 64, "out": null, "value": 424.8, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64-token prompts", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-serve-c1", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//128", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": 128, "value": 210.9, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "short chat requests, 128 generated tokens, greedy", "condition": "NVIDIA GeForce RTX 4090, dedicated single-GPU cloud host; board power not recorded; one run per level.", "dev_build": "Qwen3.6-35B-A3B earlier serve build (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-serve-c4", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//128", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": 128, "value": 377.6, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "short chat requests, 128 generated tokens, greedy", "condition": "NVIDIA GeForce RTX 4090, dedicated single-GPU cloud host; board power not recorded; one run per level.", "dev_build": "Qwen3.6-35B-A3B earlier serve build (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-serve-c8", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//128", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": 128, "value": 417.5, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "short chat requests, 128 generated tokens, greedy", "condition": "NVIDIA GeForce RTX 4090, dedicated single-GPU cloud host; board power not recorded; one run per level.", "dev_build": "Qwen3.6-35B-A3B earlier serve build (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-3090-qwen35-0.8b-512", "cell": "nvidia-cuda/rtx-3090/qwen35-0.8b/decode/512/", "axis": "decode", "model": "qwen35-0.8b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 795.7, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-3090-qwen35-2b-512", "cell": "nvidia-cuda/rtx-3090/qwen35-2b/decode/512/", "axis": "decode", "model": "qwen35-2b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 435.9, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-3090-qwen35-4b-512", "cell": "nvidia-cuda/rtx-3090/qwen35-4b/decode/512/", "axis": "decode", "model": "qwen35-4b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 208.2, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-3090-qwen35-9b-512", "cell": "nvidia-cuda/rtx-3090/qwen35-9b/decode/512/", "axis": "decode", "model": "qwen35-9b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 117.0, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen35-0.8b-512", "cell": "nvidia-cuda/rtx-4090/qwen35-0.8b/decode/512/", "axis": "decode", "model": "qwen35-0.8b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 1008.4, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen35-2b-512", "cell": "nvidia-cuda/rtx-4090/qwen35-2b/decode/512/", "axis": "decode", "model": "qwen35-2b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 548.4, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen35-4b-512", "cell": "nvidia-cuda/rtx-4090/qwen35-4b/decode/512/", "axis": "decode", "model": "qwen35-4b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 247.5, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen35-9b-512", "cell": "nvidia-cuda/rtx-4090/qwen35-9b/decode/512/", "axis": "decode", "model": "qwen35-9b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 154.9, "unit": "tok/s", "runs": 2, "min": 154.6, "max": 155.2, "spread_pct": 0.39, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; runs on separate pods; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-5090-qwen35-9b-512", "cell": "nvidia-cuda/rtx-5090/qwen35-9b/decode/512/", "axis": "decode", "model": "qwen35-9b", "device": "rtx-5090", "target": "nvidia-cuda", "in": 512, "out": null, "value": 242.4, "unit": "tok/s", "runs": 3, "min": 242.2, "max": 243.4, "spread_pct": 0.5, "bound": "exact", "build": "development", "measured": "2026-10-13", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "512-token prompt (token ids), batch 1", "condition": "NVIDIA GeForce RTX 5090; dedicated single-GPU cloud host; board power not recorded; runs on separate pods; no quiet gate.", "dev_build": "Qwen previous build cf4687d9 (2026-10-13; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen36-35b-a3b-7680", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/decode/7680/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 7680, "out": 64, "value": 251.36, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "7,680-token context, 64 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-rtx-4090-qwen36-35b-a3b-16384", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/decode/16384/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 16384, "out": 64, "value": 226.68, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "16,384-token context, 64 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-7680", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/7680/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 7680, "out": 64, "value": 259.11, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "7,680-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-16384", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/16384/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 16384, "out": 64, "value": 251.21, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "16,384-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen36-35b-a3b-7680", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen36-35b-a3b/decode/7680/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 7680, "out": 64, "value": 179.21, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "7,680-token context, 64 generated tokens, batch 1", "condition": "NVIDIA A100 80GB; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen36-35b-a3b-16384", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen36-35b-a3b/decode/16384/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 16384, "out": 64, "value": 169.79, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "16,384-token context, 64 generated tokens, batch 1", "condition": "NVIDIA A100 80GB; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9 (2026-10-14; decode kernels source-identical to the shipped package, recompiled)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-49152", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/49152/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 49152, "out": 64, "value": 218.95, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "49,152-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9, long-context variant (2026-10-14)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-64512", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/64512/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 64512, "out": 64, "value": 206.26, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64,512-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9, long-context variant (2026-10-14)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-131072", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/131072/64", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 131072, "out": 64, "value": 172.43, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "131,072-token context, 64 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g9, long-context variant (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-4090-qwen36-35b-a3b-real114-ar", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 276.6, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-4090-qwen36-35b-a3b-real114-mtp", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 373.89, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-real114-ar", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 284.11, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-h100-80gb-hbm3-qwen36-35b-a3b-real114-mtp", "cell": "nvidia-cuda/h100-80gb-hbm3/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 357.16, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA H100 80GB HBM3 (SXM); dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen36-35b-a3b-real114-ar", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 183.96, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA A100 80GB; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-a100-80gb-sxm4-qwen36-35b-a3b-real114-mtp", "cell": "nvidia-cuda/a100-80gb-sxm4/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "a100-80gb-sxm4", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 228.05, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA A100 80GB; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-3090-qwen36-35b-a3b-real114-ar", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 213.45, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-3090-qwen36-35b-a3b-real114-mtp", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/decode/114/256", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 114, "out": 256, "value": 251.74, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "114-token real math prompt, 256 generated tokens, batch 1", "condition": "NVIDIA GeForce RTX 3090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g13 (2026-10-14)" }, { "id": "cuda-dev-decode-rtx-5090-qwen36-35b-a3b-short", "cell": "nvidia-cuda/rtx-5090/qwen36-35b-a3b/decode//", "axis": "decode", "model": "qwen36-35b-a3b", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 390.8, "unit": "tok/s", "runs": 3, "min": 384.1, "max": 392.8, "spread_pct": 2.23, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "short prompt, batch 1", "condition": "NVIDIA GeForce RTX 5090; dedicated single-GPU cloud host; board power not recorded; runs on separate pods; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build with the router change (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-batch-c1", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 255.78, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "batched decode", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g11 (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-batch-c2", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 391.29, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "batched decode", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g11 (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-batch-c4", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 557.66, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "batched decode", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g11 (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-4090-qwen36-35b-a3b-batch-c8", "cell": "nvidia-cuda/rtx-4090/qwen36-35b-a3b/throughput//", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 625.47, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "batched decode", "condition": "NVIDIA GeForce RTX 4090; dedicated single-GPU cloud host; board power not recorded; no quiet gate.", "dev_build": "Qwen3.6-35B-A3B previous build g11 (2026-10-14)" }, { "id": "cuda-dev-throughput-rtx-3090-qwen36-35b-a3b-c12", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/throughput/64/", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 64, "out": null, "value": 411.7, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64-token prompts", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-throughput-rtx-3090-qwen36-35b-a3b-c16", "cell": "nvidia-cuda/rtx-3090/qwen36-35b-a3b/throughput/64/", "axis": "throughput", "model": "qwen36-35b-a3b", "device": "rtx-3090", "target": "nvidia-cuda", "in": 64, "out": null, "value": 416.8, "unit": "tok/s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-06", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "64-token prompts", "condition": "NVIDIA GeForce RTX 3090 (24 GB) in the owner's workstation at a 420 W limit, not a cloud host; other GPU activity not recorded.", "dev_build": "Qwen3.6-35B-A3B earlier build g11 (2026-10-06)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.5, "unit": "GB", "runs": 3, "min": 16.5, "max": 16.5, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.3, "unit": "GB", "runs": 5, "min": 16.3, "max": 16.3, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 29.29, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.36, "unit": "GB", "runs": 3, "min": 6.97, "max": 7.36, "spread_pct": 5.3, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3536, "unit": "s/step", "runs": 5, "min": 0.35276, "max": 0.35384, "spread_pct": 0.31, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.24, "unit": "GB", "runs": 5, "min": 15.14, "max": 15.24, "spread_pct": 0.66, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 27.96, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3537, "unit": "s/step", "runs": 5, "min": 0.35354, "max": 0.35387, "spread_pct": 0.09, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.24, "unit": "GB", "runs": 5, "min": 15.18, "max": 15.24, "spread_pct": 0.39, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 24.72, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 2.1214, "unit": "s/step", "runs": 3, "min": 2.11988, "max": 2.12149, "spread_pct": 0.08, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.72, "unit": "GB", "runs": 3, "min": 13.72, "max": 13.72, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 106.92, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 1.1462, "unit": "s/step", "runs": 3, "min": 1.14597, "max": 1.14759, "spread_pct": 0.14, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.72, "unit": "GB", "runs": 3, "min": 13.72, "max": 13.72, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 68.32, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 1.141, "unit": "s/step", "runs": 3, "min": 1.14037, "max": 1.14258, "spread_pct": 0.19, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.61, "unit": "GB", "runs": 3, "min": 10.61, "max": 10.61, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 71.61, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.7447, "unit": "s/step", "runs": 3, "min": 0.74394, "max": 0.74521, "spread_pct": 0.17, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.72, "unit": "GB", "runs": 3, "min": 13.72, "max": 13.72, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 50.65, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-step-rtx-a5000-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.7431, "unit": "s/step", "runs": 3, "min": 0.74266, "max": 0.74309, "spread_pct": 0.06, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-a5000-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.72, "unit": "GB", "runs": 3, "min": 13.72, "max": 13.72, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-a5000-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-a5000/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-a5000", "target": "nvidia-cuda", "in": null, "out": null, "value": 51.99, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA RTX A5000 (24 GB); 230 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 32.33, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 22.81, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 33.07, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 23.95, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 19.45, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 76.69, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 49.03, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 59.95, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 38.41, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-3090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-3090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 37.98, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 20.77, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.4, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 26.4, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.62, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 12.86, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 49.01, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 31.05, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 41.52, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 25.16, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-4090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-4090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 24.84, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 20.16, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.56, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.97, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 20.19, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.29, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 44.6, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 30.19, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 33.9, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 27.29, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-firstimg-rtx-5090-sd3.5-large-ultrafast", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 27.28, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-3090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-3090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-3090", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.61, "unit": "GB", "runs": 3, "min": 16.61, "max": 16.61, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-14", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 3090 (24 GB, sm_86); power 350 W, uncapped; driver 580.126.20; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-4090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-4090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-4090", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.88, "unit": "GB", "runs": 3, "min": 16.88, "max": 16.88, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 4090 (24 GB, sm_89); power 450 W, uncapped; driver 580.159.04; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-5090-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-5090/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.11, "unit": "GB", "runs": 3, "min": 17.11, "max": 17.11, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-vram-rtx-5090-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-5090/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.03, "unit": "GB", "runs": 3, "min": 14.03, "max": 14.03, "spread_pct": 0.0, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "sd35-redcar-1024 (\"A red colored car.\", 1024x1024, 40 steps, seed 42)", "condition": "NVIDIA GeForce RTX 5090 (32 GB, sm_120); power 575 W, uncapped; driver 590.48.01; dedicated single-GPU cloud host; no quiet gate.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-dev-render-rtx-5090-wan21-t2v-1.3b-balanced", "cell": "nvidia-cuda/rtx-5090/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-5090", "target": "nvidia-cuda", "in": null, "out": null, "value": 118.1, "unit": "s", "runs": 2, "min": 118.01, "max": 118.231, "spread_pct": 0.19, "bound": "exact", "build": "development", "measured": "2026-10-10", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 6, seed 7", "condition": "NVIDIA GeForce RTX 5090; 575 W, uncapped; driver 570.195.03; dedicated single-GPU cloud host; two runs.", "dev_build": "WAN 2.1 1.3B earlier build (2026-10-10)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-quality-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 8.027, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-quality-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 407.35, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-vram-rtx-3080-ti-wan21-t2v-1.3b-quality-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/vram//", "axis": "vram", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 6.88, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-balanced-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 6.648, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-balanced-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 338.43, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-vram-rtx-3080-ti-wan21-t2v-1.3b-balanced-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/vram//", "axis": "vram", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 6.88, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-fast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 5.055, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-fast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 258.77, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-vram-rtx-3080-ti-wan21-t2v-1.3b-fast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/vram//", "axis": "vram", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 6.88, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-ultrafast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 3.455, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-ultrafast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 178.81, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-vram-rtx-3080-ti-wan21-t2v-1.3b-ultrafast-480p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/vram//", "axis": "vram", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 6.88, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "832x480, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-quality-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 35.86, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-vram-rtx-3080-ti-wan21-t2v-1.3b-quality-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/vram//", "axis": "vram", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.16, "unit": "GB", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-balanced-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 29.086, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-balanced-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 1473.8, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-fast-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 20.122, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-fast-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 1025.5, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-step-rtx-3080-ti-wan21-t2v-1.3b-ultrafast-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/step_time//", "axis": "step_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 11.636, "unit": "s/step", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-dev-render-rtx-3080-ti-wan21-t2v-1.3b-ultrafast-720p", "cell": "nvidia-cuda/rtx-3080-ti/wan21-t2v-1.3b/render_time//", "axis": "render_time", "model": "wan21-t2v-1.3b", "device": "rtx-3080-ti", "target": "nvidia-cuda", "in": null, "out": null, "value": 601.2, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-22", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": "1280x720, 81 frames, 50 steps, guidance 5.0, seed 42", "condition": "NVIDIA GeForce RTX 3080 Ti (12 GB), dedicated to the run, 400 W limit, driver 591.86; 1 run(s).", "dev_build": "WAN 2.1 1.3B research build 056e44cb (2026-10-22)" }, { "id": "cuda-step-a40-sd3.5-medium-quality", "cell": "nvidia-cuda/a40/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.6126, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-medium-quality", "cell": "nvidia-cuda/a40/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.61, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-medium-balanced", "cell": "nvidia-cuda/a40/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3584, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-medium-balanced", "cell": "nvidia-cuda/a40/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.42, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-medium-low-memory", "cell": "nvidia-cuda/a40/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3562, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-medium-low-memory", "cell": "nvidia-cuda/a40/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.47, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-medium-fast", "cell": "nvidia-cuda/a40/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2888, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-medium-fast", "cell": "nvidia-cuda/a40/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.35, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2892, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.35, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-large-quality", "cell": "nvidia-cuda/a40/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 1.6789, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-large-quality", "cell": "nvidia-cuda/a40/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-large-balanced", "cell": "nvidia-cuda/a40/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.8647, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-large-balanced", "cell": "nvidia-cuda/a40/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-large-low-memory", "cell": "nvidia-cuda/a40/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.8632, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-large-low-memory", "cell": "nvidia-cuda/a40/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.66, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-large-fast", "cell": "nvidia-cuda/a40/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.59, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-large-fast", "cell": "nvidia-cuda/a40/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-a40-sd3.5-large-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.5894, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-a40-sd3.5-large-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4265, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.6, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2516, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.41, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.25, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.07, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2127, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.34, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2132, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.34, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 1.1938, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.6051, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.78, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.6016, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-rtx-pro-4500-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.66, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-rtx-pro-4500-sd3.5-large-fast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4686, "unit": "s/step", "runs": 2, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-medium-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.3622, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-medium-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.02, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-medium-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2189, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-medium-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.3, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-medium-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.2184, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-medium-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 7.48, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-medium-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1933, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-medium-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.76, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/step_time//", "axis": "step_time", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.1933, "unit": "s/step", "runs": 5, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/vram//", "axis": "vram", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.76, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-large-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.9291, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-large-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.98, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-large-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.5052, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-large-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.98, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-large-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.5044, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-large-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 10.87, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-large-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.4219, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-large-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.98, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-step-h100-80gb-hbm3-sd3.5-large-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/step_time//", "axis": "step_time", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 0.422, "unit": "s/step", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; per-step median of the denoise loop; board power as recorded in the sweep.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-vram-h100-80gb-hbm3-sd3.5-large-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/vram//", "axis": "vram", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 13.98, "unit": "GB", "runs": 3, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; peak device memory in use during the render.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-medium-quality", "cell": "nvidia-cuda/a40/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 35.0, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-medium-balanced", "cell": "nvidia-cuda/a40/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 24.77, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-medium-low-memory", "cell": "nvidia-cuda/a40/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 29.29, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-medium-fast", "cell": "nvidia-cuda/a40/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 22.21, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 21.77, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-large-quality", "cell": "nvidia-cuda/a40/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 88.01, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-large-balanced", "cell": "nvidia-cuda/a40/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 55.51, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-large-low-memory", "cell": "nvidia-cuda/a40/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 60.67, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-large-fast", "cell": "nvidia-cuda/a40/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 45.16, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-a40-sd3.5-large-ultrafast", "cell": "nvidia-cuda/a40/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "a40", "target": "nvidia-cuda", "in": null, "out": null, "value": 45.0, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-medium-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 23.68, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-medium-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 16.73, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-medium-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 26.86, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-medium-fast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 15.23, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 14.98, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-large-quality", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 60.51, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-large-balanced", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 36.69, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-rtx-pro-4500-sd3.5-large-low-memory", "cell": "nvidia-cuda/rtx-pro-4500/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "rtx-pro-4500", "target": "nvidia-cuda", "in": null, "out": null, "value": 44.79, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-medium-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 24.56, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-medium-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 19.05, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-medium-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 23.82, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-medium-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.54, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-medium-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-medium/e2e//", "axis": "e2e", "model": "sd3.5-medium", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 17.7, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-large-quality", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 56.36, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-large-balanced", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 38.9, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-large-low-memory", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 44.04, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-large-fast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 36.04, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" }, { "id": "cuda-e2e-h100-80gb-hbm3-sd3.5-large-ultrafast", "cell": "nvidia-cuda/h100-80gb-hbm3/sd3.5-large/e2e//", "axis": "e2e", "model": "sd3.5-large", "device": "h100-80gb-hbm3", "target": "nvidia-cuda", "in": null, "out": null, "value": 35.27, "unit": "s", "runs": 1, "min": null, "max": null, "spread_pct": null, "bound": "exact", "build": "development", "measured": "2026-10-03", "kind": "development", "carried_from": null, "kernel_digest": null, "protocol": "cuda-own-devbuild-v1", "prompt_set": null, "condition": "Dedicated single-GPU cloud host; one clean cold render, load+denoise end to end.", "dev_build": "SD3.5 sweep build (Oct 3 and 14)" } ] }