Veizik for NVIDIAIn preparation

Image generation on your own NVIDIA GPU, on supported NVIDIA generations: Ampere, Ada Lovelace, Hopper, Blackwell.

Whose work, which task.

Developers and small operators who generate images repeatedly on NVIDIA GPUs and want one binary without a framework install.

What comes back.

Output appears here on the day the build that produces it is public. Nothing drawn by an unpublished build is shown as a result you could reproduce.

Where it runs, and what.

• Linux x86-64 with the NVIDIA driver.

FLUX.1 — Not published yet. The engine runs it today; no download is open.; Stable Diffusion 3.5 — Not published yet. It follows FLUX.1.. Not in the public catalogue: a model is listed only once it has a measured row of its own.

From the package to a first answer.

Register this device. It opens your browser so you can approve the machine from your account, and prints the link as well in case it cannot. One activation per device -- not per model. The licence for this release is published with the release.

Nothing is downloadable yet. On the day the release is public this section carries the package, its checksum and the first commands; until then the page says so rather than offering something nobody can fetch.

The CUDA build is being prepared. The engine runs on these GPUs; no NVIDIA measurements are published yet, and what is not ready is the packaged, signed download.

Measured on this build.

Published when a row exists. A measurement is shown here only once it has a row of its own on the board, with its method and its spread.

Every row, and how to reproduce it.

Every published row names the device, the model, the token counts, the other engine's version and settings, how each side was timed, and the spread. Rows taken on earlier builds of the same cores are marked. The board →

$ veizik bench qwen2.5-7b