NVIDIA GeForce RTX 5090Available
- Vendor
- NVIDIA
- Backend
- CUDA
- Platform
- Linux on x86-64
- Generation
- Blackwell
- Compute capability
- sm_120
- Availability
- Available
Hardware
NVIDIA GeForce RTX 5090
Veizik on NVIDIA GeForce RTX 5090 — NVIDIA · CUDA · sm_120.
Overview
What this device is.
On NVIDIA GeForce RTX 5090, 2 model families are verified end to end on a published build. Evidence and availability are separate: “verified” is a first-hand run on this exact device, not a download, and “available” never means verified on your hardware.
Engines
The engine that runs here.
Veizik native engineAvailable
- Backend
- CUDA
- Architecture
- sm_120 (one kernel set per architecture)
- Model path
- dep0 — pure native, no Python runtime
- Families carried
- FLUX.1, Stable Diffusion 3.5, Wan 2.1 · Wan 2.2, Qwen3.5, Qwen3.6
One engine runs every model family on this backend. Each GPU architecture is carried as its own native kernel set; a family is “verified” on a device only when it has been run there end to end.
Models
What runs on this device.
Pick a device, get a straight answer — which model families are verified, carried by architecture, or in preparation.
●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.
| Model family | Evidence on this device | Availability |
|---|---|---|
| FLUX.1 Image | ○Expected, untested Expected, untested | Available |
| Stable Diffusion 3.5 Image | ○Expected, untested Expected, untested | Available |
| Wan 2.1 · Wan 2.2 Video | ○Expected, untested Expected, untested | Available |
| Qwen3.5 Language | ●Verified Verified | Available |
| Qwen3.6 Language | ●Verified Verified | Available |
Stable Diffusion 3.5: Powered by Stability AI
Performance
Measured on this exact device.
Decode, tokens per second Measured here
Generated tokens per second, greedy, one request, through veizik run. Median of the runs shown (hover for the range).
| Model | Mode | Workload | tok/s |
|---|---|---|---|
| Qwen3.5-0.8B | balanced | 21 / 128 tok | 1,529 tok/s · 2 runs · range 1,528–1,531 |
| Qwen3.5-0.8B | balanced | 21 / 1024 tok | 1,676 tok/s · single run |
| Qwen3.5-0.8B | fast | 21 / 128 tok | 1,531 tok/s · 2 runs · range 1,530–1,532 |
| Qwen3.5-0.8B | quality | 21 / 128 tok | 1,438 tok/s · 2 runs · range 1,437–1,438 |
| Qwen3.5-2B | balanced | 21 / 128 tok | 1,138 tok/s · 2 runs · range 1,138–1,138 |
| Qwen3.5-2B | balanced | 21 / 1024 tok | 1,107 tok/s · single run |
| Qwen3.5-2B | fast | 21 / 128 tok | 1,137 tok/s · 2 runs · range 1,135–1,138 |
| Qwen3.5-2B | quality | 21 / 128 tok | 1,041 tok/s · 2 runs · range 1,038–1,043 |
| Qwen3.5-4B | balanced | 21 / 128 tok | 772.2 tok/s · 2 runs · range 772.1–772.3 |
| Qwen3.5-4B | balanced | 21 / 1024 tok | 822.2 tok/s · single run |
| Qwen3.5-4B | fast | 21 / 128 tok | 771.4 tok/s · 2 runs · range 770.6–772.3 |
| Qwen3.5-4B | quality | 21 / 128 tok | 670.5 tok/s · 2 runs · range 670.4–670.5 |
| Qwen3.5-9B | balanced | 21 / 128 tok | 498.7 tok/s · 2 runs · range 498.6–498.8 |
| Qwen3.5-9B | balanced | 21 / 1024 tok | 434.6 tok/s · single run |
| Qwen3.5-9B | fast | 21 / 128 tok | 498.5 tok/s · 2 runs · range 498.4–498.6 |
| Qwen3.5-9B | quality | 21 / 128 tok | 423.4 tok/s · 2 runs · range 423.3–423.4 |
| Qwen3.6-35B-A3B | balanced | 21 / 128 tok | 449.9 tok/s · 2 runs · range 449.9–450.0 |
| Qwen3.6-35B-A3B | balanced | 21 / 1024 tok | 449.6 tok/s · single run |
| Qwen3.6-35B-A3B | fast | 21 / 128 tok | 544.7 tok/s · 2 runs · range 544.6–544.7 |
Prompt pass Measured here
Time for the prompt pass of the short fixed prompt, including the first generated token, as the engine reports it. Not a prefill throughput.
| Model | Mode | Workload | ms |
|---|---|---|---|
| Qwen3.5-0.8B | balanced | 21 / 128 tok | 2.7 ms · 2 runs · range 2.7–2.7 |
| Qwen3.5-0.8B | balanced | 21 / 1024 tok | 2.7 ms · single run |
| Qwen3.5-0.8B | fast | 21 / 128 tok | 2.7 ms · 2 runs · range 2.7–2.8 |
| Qwen3.5-0.8B | quality | 21 / 128 tok | 2.9 ms · 2 runs · range 2.9–3.0 |
| Qwen3.5-2B | balanced | 21 / 128 tok | 3.8 ms · 2 runs · range 3.8–3.8 |
| Qwen3.5-2B | balanced | 21 / 1024 tok | 3.8 ms · single run |
| Qwen3.5-2B | fast | 21 / 128 tok | 3.8 ms · 2 runs · range 3.8–3.8 |
| Qwen3.5-2B | quality | 21 / 128 tok | 4.1 ms · 2 runs · range 4.1–4.1 |
| Qwen3.5-4B | balanced | 21 / 128 tok | 6.9 ms · 2 runs · range 6.9–6.9 |
| Qwen3.5-4B | balanced | 21 / 1024 tok | 6.9 ms · single run |
| Qwen3.5-4B | fast | 21 / 128 tok | 6.9 ms · 2 runs · range 6.9–6.9 |
| Qwen3.5-4B | quality | 21 / 128 tok | 7.9 ms · 2 runs · range 7.9–7.9 |
| Qwen3.5-9B | balanced | 21 / 128 tok | 11.1 ms · 2 runs · range 11.0–11.1 |
| Qwen3.5-9B | balanced | 21 / 1024 tok | 11.1 ms · single run |
| Qwen3.5-9B | fast | 21 / 128 tok | 11.1 ms · 2 runs · range 11.1–11.1 |
| Qwen3.5-9B | quality | 21 / 128 tok | 13.0 ms · 2 runs · range 13.0–13.0 |
| Qwen3.6-35B-A3B | balanced | 21 / 128 tok | 61.3 ms · 2 runs · range 61.2–61.3 |
| Qwen3.6-35B-A3B | balanced | 21 / 1024 tok | 61.5 ms · single run |
| Qwen3.6-35B-A3B | fast | 21 / 128 tok | 61.3 ms · 2 runs · range 61.2–61.3 |
GPU memory in use Measured here
Once the weights and the decode graph are loaded — not the transient peak.
| Model | Mode | Workload | GiB |
|---|---|---|---|
| Qwen3.5-0.8B | balanced | — | 2.51 GiB · 2 runs |
| Qwen3.5-0.8B | fast | — | 2.51 GiB · 2 runs |
| Qwen3.5-0.8B | quality | — | 2.66 GiB · 2 runs |
| Qwen3.5-2B | balanced | — | 3.72 GiB · 2 runs |
| Qwen3.5-2B | fast | — | 3.72 GiB · 2 runs |
| Qwen3.5-2B | quality | — | 4.03 GiB · 2 runs |
| Qwen3.5-4B | balanced | — | 5.73 GiB · 2 runs |
| Qwen3.5-4B | fast | — | 5.73 GiB · 2 runs |
| Qwen3.5-4B | quality | — | 6.50 GiB · 2 runs |
| Qwen3.5-9B | balanced | — | 8.67 GiB · 2 runs |
| Qwen3.5-9B | fast | — | 8.67 GiB · 2 runs |
| Qwen3.5-9B | quality | — | 10.05 GiB · 2 runs |
| Qwen3.6-35B-A3B | balanced | — | 21.67 GiB · 2 runs |
| Qwen3.6-35B-A3B | fast | — | 21.74 GiB · 2 runs |
Every figure is veizik’s own measurement on this exact device, bound to the published build it was taken on. See Reproduce our benchmarks.
Compatibility
Architecture and its siblings.
●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.
| Model family | NVIDIA GeForce RTX 5090 | State |
|---|---|---|
| FLUX.1 | ○Expected, untested | Expected, untested |
| Stable Diffusion 3.5 | ○Expected, untested | Expected, untested |
| Wan 2.1 · Wan 2.2 | ○Expected, untested | Expected, untested |
| Qwen3.5 | ●Verified | Verified |
| Qwen3.6 | ●Verified | Verified |
Same architecture (sm_120), so a family verified on one carries to the others as Supported: NVIDIA RTX PRO 4500 Blackwell.