I know this one is infuriating — you're 30 minutes into a ranked match, the screen goes black for two seconds, the game crashes to desktop, and Event Viewer shows "Display driver nvlddmkm stopped responding and has successfully recovered" with Event ID 4101. Your GPU isn't dead. It's Windows giving up on it too fast.
The actual fix: extend the TDR timeout
Windows has a watchdog called TDR (Timeout Detection and Recovery). If a GPU doesn't respond to the OS within 2 seconds, Windows resets the driver. That's great for a genuinely hung card. It's terrible for games that spike load during shader compilation or a big draw call.
The fix is a registry tweak that gives the GPU more time before Windows pulls the plug. Open Notepad, paste this, save it as tdr-fix.reg, then double-click it and approve the UAC prompt:
Windows Registry Editor Version 5.00
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers]
"TdrDelay"=dword:00000008
"TdrDdiDelay"=dword:00000008Reboot. That's it. TdrDelay is the time (in seconds) Windows waits before resetting. TdrDdiDelay covers the DirectX kernel path. Setting both to 8 seconds clears the vast majority of game crashes on 20-series and 30-series Nvidia cards, plus RX 5000 and 6000 series AMD cards.
If you're on Windows 11 22H2 or later and want to be thorough, do it from an elevated Command Prompt instead so you know it applied:
reg add "HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers" /v TdrDelay /t REG_DWORD /d 8 /f
reg add "HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers" /v TdrDdiDelay /t REG_DWORD /d 8 /f
shutdown /r /t 0Why this works (and why the 2-second default exists)
The default TDR was designed in the Windows Vista era for workstation GPUs that either worked or didn't. Modern cards do a lot of legitimate work that takes longer than 2 seconds — compiling shaders on first launch, loading a 4K texture into VRAM, rebuilding a raytracing acceleration structure. When a game hangs on one of those steps, the driver is busy, not dead. Windows can't tell the difference, so it resets.
Bumping the delay to 8 seconds gives the driver room to finish. If your card was actually crashing (hardware fault, dying VRAM), you'll still get the crash — just 6 seconds later. That's the diagnostic value here: if the error persists with an 8-second delay, you have a real hardware problem, not a timing one.
Don't set it to 60 seconds, no matter what a forum post tells you. A genuinely hung GPU will lock your whole system for a minute before recovering. 8 seconds is the sweet spot. 10 if you're on integrated graphics.
Less common variations
AMD Instinct or Radeon Pro drivers with amdkmpfd errors: Same registry fix, but also check whether you have both the Adrenalin and Pro drivers installed. Having both is a mess — pick one. DDU both, then reinstall clean.
Laptop with hybrid graphics: The error often names the iGPU while the dGPU is the one hung. Check Event Viewer for igfx or IntelDisplay entries alongside the Nvidia one. If both fire, the real fix is disabling the iGPU in BIOS for gaming sessions, or setting the game's .exe to "High performance" in Windows Graphics Settings.
Crashes only on one specific game: Before touching the registry, update the game. Shader compilation bugs in newer titles (Helldivers 2 shipped with one, Star Wars Jedi: Survivor had two) trigger fake TDRs. A patch often fixes it. If it doesn't, then use the registry fix.
Overclocked card: TDR errors are the first sign your overclock is unstable but not unstable enough to crash outright. Undervolt slightly or drop 50 MHz off the core. If the errors vanish, your OC was the problem all along.
Crashes start after a Windows update: Roll back the driver with DDU, not Device Manager. Device Manager's rollback is a coin flip on modern GPUs.
Preventing this going forward
Three things that actually help:
- Cap your frame rate to your monitor's refresh. Uncapped FPS is the single biggest driver of TDR events. A game running at 400 FPS on a 144Hz panel is producing frames nobody sees and giving the driver no headroom.
- Skip the "clean install" checkbox in Nvidia/AMD installers. Use DDU in Safe Mode instead. The checkbox misses files. I've seen it leave three driver versions tangled together on one machine.
- Update chipset drivers when you update GPU drivers. Sounds unrelated. It isn't. A stale PCIe driver on a B550 board will trigger 4101 errors under load, and you'll spend a weekend blaming the GPU.
If the 8-second TdrDelay fix works, leave it in place. There's no downside for gaming, and the only thing you lose is a slightly slower recovery from a genuine hang. That's a fair trade for not crashing out of every match.