Quick answer: Increase the TDR delay to 8 seconds, clean-install the latest GPU driver, and check for overheating or a failing PSU.
What's actually happening here is that Windows has a watchdog called Timeout Detection and Recovery (TDR). It pings the GPU every couple of seconds. If the GPU doesn't answer within two seconds, Windows assumes it's hung, resets the driver, and you get a black screen followed by a BSOD with code 0x00000116 (TDR_FAILURE). The real trigger is rarely the driver itself — it's a GPU that's taking too long to finish a task. That can happen because of a bad overclock, a dusty fan, a dying power supply, or a driver bug that leaves work queued up. I've seen this most on aging GTX 970s and RX 580s running newer games with high texture settings, and on laptops with hybrid graphics where the discrete GPU is starved for power.
Don't just reinstall Windows. That wastes an afternoon and won't fix a hardware cause. Work through these steps in order.
Step-by-step fixes
-
Increase the TDR delay. This gives your GPU more time before Windows hits the reset button. Open Registry Editor as admin and go to
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers. Create a new DWORD (32-bit) value namedTdrDelayand set it to8(decimal). If you also seeTdrDdiDelay, set that to8too. Reboot. This alone fixes about 40% of cases where the crash happens under heavy load but not at idle. - Clean-install the GPU driver. Not an upgrade — a clean install. Use DDU (Display Driver Uninstaller) in Safe Mode. Boot into Safe Mode, run DDU, select your GPU vendor, and click "Clean and restart." Then install the latest driver from NVIDIA, AMD, or Intel. Skip GeForce Experience or Adrenalin's auto-overclock features for now. Those add instability.
- Check temperatures and power. Install HWiNFO64. Watch GPU hotspot temp and GPU power draw while gaming. If hotspot exceeds 105°C or power draw spikes above your PSU's rated wattage, you've found the cause. A failing PSU can't deliver clean power, so the GPU stalls and triggers TDR. Replace the PSU if it's older than 5 years or a no-name brand.
- Undo any overclock. This includes factory overclocks from the manufacturer. MSI Afterburner, EVGA Precision, and AMD Wattman all apply overclocks that seemed stable until they didn't. Reset to stock clocks and test. If the crash stops, you know the overclock was marginal.
- Disable hardware acceleration in browsers and apps. Chrome, Discord, and even some PDF readers use the GPU for rendering. A bug there can trigger TDR. In Chrome, go to Settings > System and turn off "Use hardware acceleration when available." Test for a day. If crashes stop, update those apps.
- Run a memory test. Bad RAM can corrupt the commands sent to the GPU. Run MemTest86 overnight. Even one error means you need to reseat or replace the RAM. This is rare but I've seen it twice this year on systems with mismatched DIMMs.
If those don't work
Try a different GPU if you have one. If the crash follows the card, the card is failing. If it stays with the system, the motherboard PCIe slot or chipset is suspect. Also test with a known-good PSU. And check for BIOS updates — some motherboards have PCIe power delivery bugs that vendors fix in later BIOS revisions.
For laptops, the issue is often the hybrid graphics driver. Force the discrete GPU for specific games via Windows Graphics Settings (Settings > System > Display > Graphics). Also make sure your laptop is plugged in — battery mode can throttle the GPU hard enough to cause timeouts.
Prevention
Keep your case clean. Dust kills GPUs. Blow out the fans and heatsinks every six months. And don't chase the last 5% of performance with aggressive overclocks — stability beats benchmarks every time. If you're on a laptop, use a cooling pad and keep the vents unobstructed. Most importantly, update your GPU driver only when you need to, not every time a new one drops. Read release notes for known issues. A driver from three months ago that works is better than a fresh one that crashes.