How to tell if your GPU is dying

GPUs almost never fail out of nowhere. They give warnings - sometimes for months - before they stop working entirely. Knowing the signs gives you time to act while the card is still diagnosable and potentially repairable.

The warning signs

Artifacts. Flickering pixels, coloured blocks, corrupted textures, or geometric shapes appearing on screen. This is the clearest indicator of GPU trouble and usually points to failing VRAM or an overheating core. See our full guide on GPU artifacting for the detail.

Crashes under load. The system runs fine on the desktop but crashes, freezes, or black-screens when you start a game or render. As cards degrade, they lose stability at full load first - the desktop is the last place problems show up.

Driver crashes that keep coming back. The occasional driver crash happens to everyone. A pattern of "Display driver stopped responding and has recovered" messages, or repeated driver timeouts in games, suggests the hardware is struggling rather than the software.

Fan behaviour changes. Fans suddenly running at full speed constantly, or not spinning at all under load, point to either a sensor problem or a card desperately trying to manage temperatures it can no longer control.

Performance dropping over time. If the same game at the same settings runs noticeably worse than it did six months ago - and temperatures have crept up alongside - the cooling on the card is degrading, which accelerates everything else.

Visual glitches in specific scenarios. Corruption that only appears during video playback, or in hardware-accelerated browser content, can indicate failure in specific parts of the card rather than the whole thing.

Confirm it's actually the GPU

Several other faults produce identical symptoms, so rule these out before condemning the card:

Do a clean driver reinstall. Boot into safe mode, run DDU (Display Driver Uninstaller), then install a fresh driver from Nvidia or AMD directly. Corrupted drivers mimic dying hardware closely.

Check temperatures. Run HWiNFO64 and watch GPU temperature and hot spot temperature under load. A card hitting thermal limits will throttle, stutter, and crash - and that may be fixable with cleaning and repasting rather than terminal.

Check the power supply. A degrading PSU causes crashes under GPU load that look exactly like GPU failure, because GPU load is when the system draws the most power. If your PSU is old or barely adequate for the card, it's a suspect.

Test the card in another system, or another card in yours. This is the definitive test. If the symptoms follow the card, it's the card. If they stay with the system, look at the PSU, motherboard, or RAM.

Re-seat the card and check power cables. Remove the GPU, inspect the PCIe contacts, reseat it firmly, and reconnect all power cables. Marginal connections cause intermittent faults that look like degradation.

What a dying GPU means - and what it doesn't

A failing GPU isn't always a dead GPU. What matters is which part is failing:

• Thermal degradation - dried paste, dust-packed heatsink, failing fans. Fully recoverable with a service.
• VRAM failure - causes artifacts and crashes. A single failed memory chip can often be replaced at component level.
• Power stage failure - failing MOSFETs or capacitors in the card's VRM. Diagnosable and frequently repairable.
• Core/die failure - cracked solder under the GPU die can be repaired by reballing. Actual silicon death cannot.

The economics depend on the card. A component-level repair often costs a fraction of the card's replacement value on anything mid-range or above. On older budget cards, replacement may make more sense - an honest repair shop will tell you which side of the line you're on.

When to stop and send it in

If you've ruled out drivers, temperatures, and the PSU, and the symptoms persist or are getting worse, get the card diagnosed before it fails completely. A card that still partially works is easier to diagnose - the fault can be observed and traced. We bench-test GPUs under controlled load, identify the failing component, and give you a clear repair-or-replace recommendation before any work is done.