How to Troubleshoot a Graphics Card That Keeps Crashing

Your screen goes black mid-match. The fans spin up like a jet engine, then everything comes back — sometimes. Other times you're staring at a frozen frame with a driver error notification. You reboot, it works fine for twenty minutes, then it happens again. Sound familiar?

I've dealt with this exact problem on three different machines over the past few years, and here's what I learned the hard way: a graphics card that keeps crashing is a symptom, not a diagnosis. The GPU is usually the messenger. The actual culprit could be a dying power supply, a corrupt driver, unstable RAM, or yes — a GPU that's genuinely on its way out. The trick is figuring out which one before you spend money replacing the wrong part.

This guide walks you through how to troubleshoot a graphics card that keeps crashing, in the order I actually do it — from the cheapest, most likely fixes to the ones that involve opening your wallet.

Why is my graphics card constantly crashing?

There are really only four categories of cause: software (drivers, Windows corruption), power delivery (PSU, cables, connectors), thermal (dust, dried paste), and hardware failure (VRAM, capacitors, solder joints). Everything else is a variation on these.

The problem is that all four produce almost identical symptoms. A black screen from a bad driver looks exactly like a black screen from a failing VRAM module. This is why people throw money at new GPUs when the fix was a $4 cable.

GPU crashes but PC stays on

This is the classic pattern, and it's actually good news. If your PC keeps running — fans spinning, RGB still lit, audio sometimes continuing — the crash is almost certainly in the GPU or its driver, not the whole system. A total shutdown or instant power-off points to the PSU instead.

When the display dies but the machine lives, you're looking at either a TDR event (Timeout Detection and Recovery — Windows detected the GPU stopped responding and reset the driver) or a driver-level crash. Check Event Viewer for Event ID 4101, which is Windows literally telling you "the display driver stopped responding and has recovered."

My graphics card keeps crashing when I play games

Crashing under load specifically narrows things down considerably. If it never crashes on the desktop but dies five minutes into a game, the suspects are power draw, heat, or an unstable overclock — not idle-time software bugs.

Here's what I do: I lower the power limit in MSI Afterburner by 20% and play again. If the crashes stop, the problem is power delivery or thermal throttling, not the GPU dying. If they continue, it's deeper.

How to tell if a GPU is dying

Real hardware death has a signature. It's rarely random — it follows a pattern.

• Visual artifacts that appear immediately — colored lines, checkerboard patterns, or pixel snow at boot, before Windows even loads.
• Crashes that get progressively more frequent over weeks, not days.
• Failures at idle or in the BIOS, where no driver is loaded at all. This is the biggest tell.
• Artifacts in one specific application only, then spreading to others.
• The card won't POST in a different machine.

Artifacts appearing in the BIOS screen are damning. No driver runs there — only firmware and hardware. If you see corruption before the operating system loads, you have a hardware fault, and no amount of DDU will save it.

By contrast, a card that only crashes after a Windows update, or only in one game, is almost never dying. Those are software issues wearing a hardware costume.

GPU keeps crashing not overheating

Good — rule heat out and you've eliminated one variable. Now check power. An aging PSU that delivered clean 12V rails when new can sag under transient loads a few years later. Modern GPUs pull huge short bursts well above their rated wattage, and a weak or worn PSU trips protection and the GPU drops out.

I once chased a "dying GPU" for two weeks. Turned out the 12VHPWR connector wasn't fully seated. Reseating it — a ten-second fix — solved it completely.

Step-by-step troubleshooting (in the order that actually works)

Do these in sequence. Don't skip ahead to replacing hardware until you've cleared the cheap stuff.

1. Update or roll back the driver. Sometimes the newest NVIDIA driver is the problem. Roll back to the previous version if crashes started right after an update.
2. Run DDU in Safe Mode. Display Driver Uninstaller removes every trace of the old driver. Regular uninstalls leave fragments behind that cause conflicts.
3. Repair Windows system files. Run these in PowerShell as administrator: sfc /scannow then DISM /Online /Cleanup-Image /RestoreHealth.
4. Reseat the card and power cables. Pull the GPU out, blow out the slot, push it firmly back in, click the PCIe power connectors in until they latch.
5. Monitor with HWiNFO or GPU-Z. Watch GPU temperature, hotspot temp, power draw, and clock speeds right up to the crash moment.
6. Stress test with OCCT or FurMark. Run for 15 minutes. A crash here that doesn't happen in games tells you it's stress-specific.
7. Check Event Viewer. Look under Windows Logs > System for WHEA errors (hardware) or Event ID 4101 (driver timeout).
8. Test in another PC. If it crashes there too, the card is the problem. If it's stable, your system is.

Steps 1 through 3 fix the majority of cases. If you've done all eight and the card still crashes in a different machine, you're looking at RMA or replacement.

What triggers a GPU crash dump?

A crash dump is Windows capturing the state of the GPU and driver at the moment of failure, so it can be analyzed afterward. It gets triggered when the driver stops responding to commands within the timeout window — usually around two seconds.

Windows then attempts a TDR: it resets the graphics driver. If recovery succeeds, you get a brief black screen and a notification. If it fails, you get a full crash or reboot. The dump itself lives in %LOCALAPPDATA%\CrashDumps or in the Windows minidump folder, and tools like BlueScreenView or WhoCrashed can read it to tell you which module failed.

The most common triggers: unstable overclocks, undervolts that are too aggressive, a driver bug, a power delivery hiccup, or a failing VRAM chip that can't hold data reliably. If you undervolt your GPU for temperatures, an unstable undervolt is a frequent and often-overlooked cause of random crash dumps.

Is it possible to fix a dying GPU?

Sometimes. It depends entirely on what's dying.

Problem	Fixable?	Method
Dried thermal paste	Yes	Repaste the GPU die and replace thermal pads
Overheating VRAM	Yes	New thermal pads on memory modules
Unstable overclock/undervolt	Yes	Reset to stock clocks in Afterburner
Failing VRAM chip	Sometimes	Professional reball or chip replacement
Cracked solder joints	Rarely	Professional reflow — usually not worth it
Dead core	No	Replacement only

Repasting is the one I recommend trying yourself if the card is out of warranty. Cards from 2019 through 2022 with a few years of daily gaming often have paste that's turned to chalk. Repasting dropped my own GPU's hotspot temperature by 18°C and eliminated crashes that had been happening under load for months.

Downclocking via MSI Afterburner is the pragmatic middle ground — if a card is marginal and you just need it stable until you can replace it, shaving 100-150MHz off the core often buys you months.

Why does my graphics driver keep crashing in Valorant?

Competitive shooters get blamed a lot, but the game is rarely the cause. Valorant uses anti-cheat software that runs at a low level, which conflicts with certain driver features and overlays more readily than most games.

Turn off any overlays — Discord, GeForce Experience, MSI Afterburner's OSD — and test. Then update to the latest driver or step back one version. If crashes only happen in Valorant and nowhere else, it's almost always an overlay or anti-cheat conflict, not your card.

When to stop fixing and replace

Here's my honest rule: if you've done DDU, repaired system files, reseated everything, repasted, tested in a second machine, and it still crashes there — the card is done. Stop pouring hours into it.

The math is unkind to long troubleshooting sessions. A used equivalent GPU costs a fraction of what professionals charge for board-level repair, and repaired cards have no warranty. If the card is under warranty, skip everything after step 5 and file an RMA immediately.

What I've learned from chasing these problems is that the answer is usually boring. It's a loose cable, an out-of-date driver, or paste that gave up. The dramatic "my GPU is dying" scenario is real, but it's the last thing to suspect — not the first. Test the boring explanations, and you'll save yourself both time and money.

And if you've genuinely tried everything and the card still fails? That's not a failure of your troubleshooting. That's just a GPU that lived its life.