Hardware

NVIDIA RTX A5000

Veizik on NVIDIA RTX A5000 — NVIDIA · CUDA · sm_86.

Overview

What this device is.

NVIDIA RTX A5000Available

Vendor: NVIDIA
Backend: CUDA
Platform: Linux on x86-64
Generation: Ampere
Compute capability: sm_86
Availability: Available

On NVIDIA RTX A5000, 1 model family is verified end to end on a published build, 2 more carried by the same architecture. 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.

All devices · Platform requirements · Benchmarks & method

Engines

The engine that runs here.

Veizik native engineAvailable

Backend: CUDA
Architecture: sm_86 (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	Architecture-supported Architecture-supported	Available
Qwen3.5 Language	Verified Verified	Available
Qwen3.6 Language	Architecture-supported Architecture-supported	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	20 / 128 tok	776.7 tok/s · 2 runs · range 776.2–777.1
Qwen3.5-0.8B	balanced	921 / 1024 tok	916.1 tok/s · single run
Qwen3.5-0.8B	fast	20 / 128 tok	771.9 tok/s · 2 runs · range 771.5–772.2
Qwen3.5-0.8B	quality	20 / 128 tok	714.7 tok/s · 2 runs · range 712.2–717.1
Qwen3.5-2B	balanced	20 / 128 tok	467.3 tok/s · 2 runs · range 465.8–468.9
Qwen3.5-2B	balanced	921 / 1024 tok	449.1 tok/s · single run
Qwen3.5-2B	fast	20 / 128 tok	464.6 tok/s · 2 runs · range 463.5–465.7
Qwen3.5-2B	quality	20 / 128 tok	435.2 tok/s · 2 runs · range 434.8–435.6
Qwen3.5-4B	balanced	20 / 128 tok	280.5 tok/s · 2 runs · range 279.3–281.7
Qwen3.5-4B	balanced	921 / 1024 tok	228.6 tok/s · single run
Qwen3.5-4B	fast	20 / 128 tok	278.9 tok/s · 2 runs · range 278.0–279.9
Qwen3.5-4B	quality	20 / 128 tok	256.5 tok/s · 2 runs · range 255.9–257.1
Qwen3.5-9B	balanced	20 / 128 tok	151.0 tok/s · 2 runs · range 150.8–151.1
Qwen3.5-9B	balanced	921 / 1024 tok	124.8 tok/s · single run
Qwen3.5-9B	fast	20 / 128 tok	149.1 tok/s · 2 runs · range 148.8–149.3
Qwen3.5-9B	quality	20 / 128 tok	137.8 tok/s · 2 runs · range 137.2–138.4

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	20 / 128 tok	4.9 ms · 2 runs · range 4.9–4.9
Qwen3.5-0.8B	balanced	921 / 1024 tok	28.6 ms · single run
Qwen3.5-0.8B	fast	20 / 128 tok	4.9 ms · 2 runs · range 4.9–4.9
Qwen3.5-0.8B	quality	20 / 128 tok	4.6 ms · 2 runs · range 4.6–4.7
Qwen3.5-2B	balanced	20 / 128 tok	8.4 ms · 2 runs · range 8.4–8.4
Qwen3.5-2B	balanced	921 / 1024 tok	58.5 ms · single run
Qwen3.5-2B	fast	20 / 128 tok	8.4 ms · 2 runs · range 8.4–8.4
Qwen3.5-2B	quality	20 / 128 tok	9.0 ms · 2 runs · range 9.0–9.0
Qwen3.5-4B	balanced	20 / 128 tok	17.2 ms · 2 runs · range 17.0–17.5
Qwen3.5-4B	balanced	921 / 1024 tok	154.0 ms · single run
Qwen3.5-4B	fast	20 / 128 tok	17.2 ms · 2 runs · range 16.9–17.5
Qwen3.5-4B	quality	20 / 128 tok	19.2 ms · 2 runs · range 18.9–19.4
Qwen3.5-9B	balanced	20 / 128 tok	29.0 ms · 2 runs · range 29.0–29.0
Qwen3.5-9B	balanced	921 / 1024 tok	269.9 ms · single run
Qwen3.5-9B	fast	20 / 128 tok	29.2 ms · 2 runs · range 29.2–29.2
Qwen3.5-9B	quality	20 / 128 tok	31.0 ms · 2 runs · range 30.9–31.0

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	—	2.22 GiB · 2 runs
Qwen3.5-0.8B	fast	—	2.22 GiB · 2 runs
Qwen3.5-0.8B	quality	—	2.37 GiB · 2 runs
Qwen3.5-2B	balanced	—	3.43 GiB · 2 runs
Qwen3.5-2B	fast	—	3.43 GiB · 2 runs
Qwen3.5-2B	quality	—	3.75 GiB · 2 runs
Qwen3.5-4B	balanced	—	5.44 GiB · 2 runs
Qwen3.5-4B	fast	—	5.44 GiB · 2 runs
Qwen3.5-4B	quality	—	6.22 GiB · 2 runs
Qwen3.5-9B	balanced	—	8.39 GiB · 2 runs
Qwen3.5-9B	fast	—	8.39 GiB · 2 runs
Qwen3.5-9B	quality	—	9.77 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.

Model family	NVIDIA RTX A5000	State
FLUX.1	Expected, untested	Expected, untested
Stable Diffusion 3.5	Expected, untested	Expected, untested
Wan 2.1 · Wan 2.2	Architecture-supported	Architecture-supported
Qwen3.5	Verified	Verified
Qwen3.6	Architecture-supported	Architecture-supported

Same architecture (sm_86), so a family verified on one carries to the others as Supported: NVIDIA GeForce RTX 3080 Ti, GeForce RTX 3090, NVIDIA A40.