VeizikNVIDIA, Apple Silicon and edge builds are available now.Release notes →
veiziknative inference engine
For business Sign inDownload
Hardware

What can your device run?

Choose your hardware to see available models, measured results and requirements.

Find your device

Pick your hardware.

Search by name, or start from a platform. Each card keeps availability, architecture support, verification and measured results apart.

NVIDIA GPUs

Available

Ampere · Ada Lovelace · Hopper · Blackwell · 10 devices

Linux on x86-64 · CUDA

Apple Silicon

Available

M1 · M2 · M4 · 4 devices

macOS on Apple Silicon (arm64) · Metal

Edge devices

Raspberry Pi · Jetson · Android · 5 devices

Arm CPU, NPU and edge GPU

Search by name, chip or architecture. You do not need to know a compute capability.

Filters

Architecture

Showing all 19 devices.

#find

Registry data as of 2026-11-14. Cards count model families, not variants. Memory is the vendor-listed capacity of the SKU; whether a model fits depends on its measured peak plus a reserve, not on the size of its weights.

What runs on it

See what runs on your device.

Choose a device for its model list: what is on offer, what has been run on it, and how many results exist.

Choose a device to see its model list, with availability, architecture support and measured results in separate columns.

Stable Diffusion 3.5: Powered by Stability AI

How to read this page

Four counts, kept separate.

Availability comes from the release registry, evidence from first-hand runs, results from our own measurements. One never stands in for another.

01

Available

A downloadable build on offer includes this device’s architecture. Read from the release registry; it does not say the model was run on this device.

Releases →
02

Architecture-supported

Verified end to end on another device of the same architecture. For NVIDIA that means the same compute capability: a kernel set that ran on one says nothing about another.

How we verify →
03

Verified

Run end to end on this exact device, on a published build. Not a download check and not a speed claim.

See the matrix →
04

Measured

Our own results on this exact device with a published build, each with its model, mode, workload and run count. Earlier-build figures are counted apart.

Open benchmarks →
Support is not the same as every variant, or a fit.

A family listed as available can have variants that are not in the build, and a verified family does not mean every variant was measured. Memory capacity on a card is the vendor-listed figure; whether a model fits depends on its measured peak plus a reserve, never on the size of its weights alone.

Advanced · matrix

Every device against every model family.

The full grid for readers who want to compare devices side by side. Each cell is one device and one family.

VerifiedVerifiedRun end to end on this exact device, on a published build.Architecture-supportedArchitecture-supportedVerified on another device of the same architecture; this device itself is untested.Expected, untestedExpected, untestedThe engine targets this architecture; nobody has run this model on it. No claim is made.In preparationIn preparationNo build is downloadable yet, and no figure is offered.Not offeredNot offeredThis family is not offered on this platform.UnknownUnknownNo information yet — the platform is announced with the release.
NVIDIA · CUDA10 devices × 5 families
DeviceFLUX.1Stable Diffusion 3.5Wan 2.1 · Wan 2.2Qwen3.5Qwen3.6
Ampere
NVIDIA GeForce RTX 3080 Ti sm_86Expected, untestedExpected, untestedArchitecture-supportedArchitecture-supportedArchitecture-supported
GeForce RTX 3090 sm_86Expected, untestedExpected, untestedVerifiedVerifiedVerified
NVIDIA RTX A5000 sm_86Expected, untestedExpected, untestedArchitecture-supportedVerifiedArchitecture-supported
NVIDIA A40 sm_86Expected, untestedExpected, untestedArchitecture-supportedArchitecture-supportedArchitecture-supported
NVIDIA A100 80GB PCIe sm_80Expected, untestedExpected, untestedExpected, untestedExpected, untestedExpected, untested
NVIDIA A100-SXM4-80GB sm_80Expected, untestedExpected, untestedExpected, untestedExpected, untestedExpected, untested
Ada Lovelace
NVIDIA GeForce RTX 4090 sm_89Expected, untestedExpected, untestedExpected, untestedVerifiedVerified
Hopper
NVIDIA H100 80GB HBM3 sm_90Expected, untestedExpected, untestedExpected, untestedExpected, untestedExpected, untested
Blackwell
NVIDIA GeForce RTX 5090 sm_120Expected, untestedExpected, untestedExpected, untestedVerifiedVerified
NVIDIA RTX PRO 4500 Blackwell sm_120Expected, untestedExpected, untestedExpected, untestedArchitecture-supportedArchitecture-supported
Apple Silicon · Metal4 devices × 2 families
DeviceQwen2.5Qwen3
M1
Apple M1 Ultra M1VerifiedVerified
Apple M1 Max M1VerifiedVerified
M2
Apple M2 Pro M2VerifiedVerified
M4
Apple M4 Pro M4VerifiedVerified

Evidence and availability are separate: “Verified” never means downloadable, and “Available” never means verified on your device. Edge devices are not in this matrix because the registry holds no per-board evidence for them. Families the engine is compatible with but does not offer are in the model catalogue.

Stable Diffusion 3.5: Powered by Stability AI

Requirements

What each platform needs.

NVIDIA
Platform
Linux on x86-64
Backend
CUDA
Generations
Ampere, Ada Lovelace, Hopper, Blackwell
Minimums
An NVIDIA GPU of compute capability 8.0 or newer, and an NVIDIA driver that provides libcuda.so.1. No CUDA toolkit and no Python — only the driver. The same build runs under both the CUDA 12 and CUDA 13 driver series; there is no separate CUDA-12 or CUDA-13 download. Linux x86-64, run natively (WSL is not supported).
Apple Silicon
Platform
macOS on Apple Silicon (arm64)
Backend
Metal
Generations
M1, M2, M3, M4
Minimums
Apple Silicon (M1 or newer), and macOS 13 or later. It is not tied to one macOS version: 13 is the minimum tested, and newer releases run.

Edge boards and phones. Their requirements are published with each model and on the Edge page; none are stated here until confirmed.

No bundled dependencies. The engine process is a single binary, and it links only three shared libraries: the NVIDIA driver (libcuda.so.1), the C library (libc) and the math library (libm). No CUDA runtime, no cuBLAS or cuDNN, no PyTorch and no Python — not installed, and not hidden inside the package. The package also carries a small launcher and a seal checker; those link only libc. On Apple Silicon the one dependency is Metal. You can check this yourself on the files you download, with readelf -d (or ldd) for the linked libraries, sha256sum -c against the published checksums, and the bundled seal checker.

Verify it yourselfThree commands on the files you download
bash · verify
# compare to the SHA-256 on the release page
$ sha256sum veizik-<model>-<version>-linux-x86_64.tar
# checks every sealed file and prints VZSEAL OK
$ bin/vzseal <unpacked-folder>
# lists the libraries the engine links
$ readelf -d bin/<engine> | grep NEEDED

The third command prints the dependency list above straight from the binary. To reproduce the measured figures as well, follow Reproduce our benchmarks.

Your hardware

Check before you download.

Not on the list?

Not on the list means untested, and no claim is made for hardware we have not run. Check your GPU’s compute capability against the architectures above before you download.

Check my hardwareBrowse models