GPU Driver Crashes in Games: Fix TDR and Display Errors

You're 20 minutes into a ranked match, the screen goes black for two seconds, the game crashes to desktop, and Event Viewer shows something like Display driver nvlddmkm stopped responding and has successfully recovered. Or worse — a blue screen stamped VIDEO_TDR_FAILURE (0x00000116). The GPU itself is usually fine. What's actually happening is Windows lost patience with your graphics driver.

Cause 1: TDR timeout — Windows killed the driver before it finished

Windows has a watchdog called TDR (Timeout Detection and Recovery). If a single GPU command takes more than 2 seconds to complete (the default is 2, but it can effectively trigger sooner under load spikes), Windows assumes the GPU is hung and resets the driver. The screen flickers black, the game dies, but Windows itself survives. That's the "stopped responding and has successfully recovered" message — it's Windows telling you it rescued itself.

The reason this hits games and not your desktop is that games are the only workload that pushes the GPU into long-running compute passes and heavy shader compilation. A texture upload during a scene transition, a spike in VRAM pressure, or a driver bug in a specific DX12 codepath can blow past that 2-second threshold. This is extremely common on fresh driver releases — especially NVIDIA's 545.xx and 546.xx branches and AMD's Adrenalin 23.11 and later, where day-one game drivers shipped with TDR regressions.

Before touching anything else, confirm it's actually a TDR hit. Open Event Viewer (eventvwr.msc), go to Windows Logs → System, and filter by source Display or nvlddmkm. You'll see entries timestamped exactly when the game died. If you see those, this is your cause.

The real fix isn't to disable TDR — that's a band-aid that hides the actual problem and can lock up your whole system on a genuinely hung GPU. The real fix is to identify which component is stalling, and the fastest way to do that is a clean driver reinstall.

1. Download DDU (Display Driver Uninstaller) from Wagnardsoft.
2. Boot into Safe Mode (hold Shift while clicking Restart → Troubleshoot → Advanced → Startup Settings).
3. Run DDU, select your GPU vendor, click "Clean and restart."
4. Install the driver version immediately before your current one. Not the latest. Driver regressions are real and frequent.

If the crash follows a specific game, check its subreddit or Steam forums for that driver version. Someone has already confirmed which one broke it.

Cause 2: Overclocked or undervolted GPU (including factory OC cards)

Every factory-overclocked card ships with a profile that's stable in benchmarks but marginal under real game workloads. 15 minutes of Cyberpunk 2077 at 4K with ray tracing is a different beast than a 60-second Time Spy run. When the GPU core or memory can't hold its clocks, the driver hits a hardware error and recovers by resetting — same symptoms as TDR, same Event Viewer entries, but the cause is silicon, not software.

Here's the test: install MSI Afterburner, drop the Core Clock offset by -100 MHz and the Memory Clock by -200 MHz, and play the crashing game for an hour. If it stops crashing, you've found your problem. The undervolt crowd sees this constantly — a curve that's stable in Heaven benchmark is often unstable in a game that hits different power states.

Same logic applies to undervolting. A -100mV offset that seemed fine for a week can crash the moment ambient temps rise in summer or a driver update changes how the GPU boosts. If you undervolt, dial it back by 25mV and retest before blaming the driver.

Reference: NVIDIA's own documentation states that TDR errors can be caused by GPU instability, and recommends testing at stock clocks before submitting an RMA. AMD's equivalent guidance is buried in their Adrenalin known-issues pages.

Cause 3: PCIe link instability, riser cables, and power delivery

This is the one people skip, and it's the one that bites hardest. A GPU that loses its PCIe link for even a microsecond — from a marginal riser cable, a dirty slot, or a PSU that can't hold 12V under transient spikes — will throw the same driver-crash errors.

You're most likely to see this if you're running a vertical GPU mount, an ITX build with a riser, or a 30-series/40-series NVIDIA card on a PSU under 750W. The transient spikes on a 3080 can hit 500W for microseconds, and a cheap PSU's OCP trips or the 12V rail sags. The GPU errors out, driver resets, game crashes.

Diagnostic steps, in order:

• Reseat the GPU. Pull it, check the gold fingers for dust or discoloration, push it back firmly until the retention clip clicks.
• If you're on a riser, test with the card in the slot directly. Riser cables are the single most common cause of "random" GPU crashes on builds that otherwise look fine.
• Check your PCIe link width and speed with GPU-Z under load. If it drops from x16 4.0 to x8 or x4 mid-game, that's a link training failure — often the riser or slot.
• Run HWiNFO64 and log the +12V rail during a crash. If you see it dip below 11.4V, the PSU is your problem.

Also check your motherboard's BIOS. Some Z690 and X670 boards shipped with PCIe 5.0 auto-negotiation bugs that cause intermittent link drops on 4.0 cards. Force the slot to Gen4 in BIOS if you're on one of those platforms.

Quick reference: symptom → likely cause → fix

One last thing, because it comes up constantly: disabling TDR entirely via the registry key HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers\TdrDelay does not fix this. It masks it. Your GPU is still hanging — you've just told Windows to wait longer before killing the driver. On a genuinely bad hang, you'll get a full system lockup instead of a recoverable crash, which is worse. Set TdrDelay back to defaults and fix the actual cause.