Hardware verification for the used GPU market
Prove the card works before you ask a stranger to trust you.
A used GPU listing is a claim with nothing behind it. GPU Cert runs a fixed test protocol against the card in your machine, then issues a signed certificate at a public URL. The buyer checks it themselves. No account needed to run it.
What gets tested
Stress test
Five minutes of sustained compute load through a hand-written Vulkan kernel, logging temperature and clocks throughout.
Catches cards that overheat, never thermally settle, or throttle hard under load.
VRAM pattern test
A bit-pattern write and verify sweep across the card's active memory, ten minutes, derived from memtest_vulkan.
Catches the memory damage heavy mining leaves behind. Telemetry alone cannot see it. One error fails the card.
Render integrity test
Renders a deterministic shader and checks every pixel of every frame against the expected output, exactly, with no tolerance.
Catches rasterizer and shader-core defects that produce wrong pixels rather than no pixels, which neither test above reads back far enough to notice.
How it works
- Download and run gpu-cert.exeA console app. It reads the card's identity, runs the three tests, and submits the result.
- Your browser opens the finished certificateSigned server-side and bound to that specific card's fingerprint, at a public URL.
- Put the link in your listingAnyone can open it and check the result without installing anything or taking your word for it.
- Optional: put your name on itThe certificate is public and verifiable without an account. To attach one to your name, so it appears in your register and you can build a track record, you need an account with a confirmed email. Connect it to the app once and every later run files itself; certificates already made can be attached afterwards.
Why a buyer should believe it
The certificate is signed with GPU Cert's key when it is issued, not generated in the browser showing it to you. Anyone can re-check that signature on the verification page, or against GPU Cert's published signing key without trusting this site at all.
There are plenty of good diagnostic tools already. None of them produce something a stranger can check.