$ SOLVED // it worked
Solved Windows NVIDIA RTX 5090 (applies to any modern GPU) moderate

GPU artifacts only occasionally — how to reproduce it on demand for a warranty claim

Symptom Checkerboard patterns, snow, or random glitches appear once every few days under normal use, but never when a technician looks — so the warranty/insurance claim goes nowhere.

Updated 2026-04-09 Verified fix

A dying GPU that only artifacts once every few days is a warranty nightmare. Insurance adjusters and warranty technicians don’t need a “natural” failure — they just need to see the hardware fail under load. A card that glitches casually once in a while will never fail while a technician is watching, so the claim stalls indefinitely.

The fix is to stop waiting for a random glitch and provoke the fault on demand, on camera, at stock settings.

Provoke it with the right stress test

  • OCCT — VRAM test. The gold standard for catching a dying card. It fills VRAM and checks for errors. Set memory usage to 95% — not 99%, because Windows needs some headroom. Artifacting is often one specific faulty memory chip out of the 16+ on the board; filling memory forces the GPU to actually touch the bad region. If the card is failing, OCCT starts counting errors in a red box — a smoking gun for the adjuster.
  • FurMark (“the donut of death”). Pushes power draw to the absolute maximum. A weak VRM or power-delivery fault will crash the card or show “snow” (tiny white dots) within minutes.
  • Unigine Superposition (8K Optimized). Forces the card to use almost all of its VRAM in a real workload.

Heat-soak

Most hardware failures only show up once the card reaches its steady-state temperature, ~75–85°C. Let the stress test run 20–30 minutes before concluding anything. A card that’s clean cold can still fail hot.

Record evidence that holds up

The video needs to prove it’s your card and the failure is real:

  1. Start on the chassis — film the serial-number sticker on the back of the PC.
  2. One take, no edits — keep the camera rolling from the serial number to the monitor.
  3. Show stock settings — prove you aren’t manually overclocking, or they’ll blame you for the instability.
  4. Capture the failure — point at the screen and narrate (“checkerboard patterns under 90% load”).

If it won’t artifact

Run OCCT for an hour and FurMark for an hour. Zero errors means the card isn’t broken. In that case the “artifacting” is probably one of the look-alikes:

  • A bad DisplayPort/HDMI cable — can mimic a dying GPU exactly.
  • Monitor ghosting/panel failure — the display, not the card.
  • Driver conflicts — a clean reinstall with DDU (Display Driver Uninstaller) fixes a lot of “glitches” people mistake for hardware death.

Intermittent issues that don’t show up on a stress test don’t get paid out. A fault you can trigger on command does.