GPU Compute Driver Timeout on Windows – Fix It Now

Quick answer

Disable TDR by setting TdrLevel to 0 in registry at HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers, then reboot. If that doesn't stick, update your GPU driver fresh from Nvidia or AMD's site.

What's happening here?

Your GPU (graphics card) stops responding to Windows for more than 2 seconds. Microsoft built in a timeout detection and recovery (TDR) feature that kills the driver to prevent a full system freeze. Problem is, compute workloads—like rendering, mining, or AI training—push the card harder than games. The driver can get busy and miss the heartbeat signal. The error code you'll see is often 0x887A0006 or just a generic "Display driver stopped responding." I've seen this on RTX 3060s, RX 6700 XTs, and even old GTX 1080s. The fix is the same.

Fix steps – in order of what works

1. Update your GPU driver – Go to Nvidia or AMD's website, download the latest Studio driver (not Game Ready). Game Ready drivers trade stability for performance in games. Studio drivers are better for compute. Do a clean install using DDU (Display Driver Uninstaller) in safe mode. Don't bother with Windows Update – it often gives you old drivers.
2. Disable TDR via Registry – Open Regedit. Navigate to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers. Right-click > New > DWORD (32-bit). Name it TdrLevel, set value to 0. That turns off the 2-second timeout. Reboot. If you still get timeouts, increase TdrDelay instead (set to 10 as a DWORD). That gives your GPU 10 seconds instead of 2. Don't set TdrLevel to 0 if you're using the PC for gaming – it can cause black screens that never recover.
3. Check your power supply – A weak or failing PSU causes voltage drops. The GPU drops out, driver times out. If your PSU is under 600W for a mid-range card, or you're using daisy-chained power cables (one cable with two connectors), split them into separate cables from the PSU. I've fixed this exact issue on a system with a 550W PSU and an RTX 3070.
4. Lower GPU clock speed – Use MSI Afterburner. Drop core clock by 50 MHz, memory by 100 MHz. Some cards come overclocked from factory and aren't stable under compute load. This is especially common with factory-overclocked cards and high ambient temps.

Alternative fixes if the main ones fail

• Disable hardware acceleration in your browser and Discord. Go to Settings > System > Display > Graphics settings > toggle hardware-accelerated GPU scheduling off. This frees up GPU time for compute tasks. I've seen Chrome trigger timeouts on its own.
• Switch to Windows Graphics Performance Preference – For the specific app that crashes, go to Settings > System > Display > Graphics > Add your app, then set it to "High performance." This stops Windows from sharing GPU resources badly.
• Reinstall chipset drivers – Go to your motherboard manufacturer's site (e.g., Asus, MSI) and get the chipset driver for your CPU (AMD or Intel). Outdated chipset drivers sometimes cause PCIe timing issues that look like GPU timeouts. I've seen this on X570 boards.

Prevention tip

Keep your GPU under 80°C during compute loads. Use HWMonitor to check. Over 85°C? Clean your fans, repaste the GPU if you're brave, or add more case fans. Heat is the number one cause of driver instability. Also, don't run compute workloads on a GPU that's also driving 3 monitors. The extra display overhead makes the card busier and more likely to miss the TDR heartbeat. One monitor is fine. Two is okay. Three? Expect problems.

If you're still crashing after all this, your GPU might have a hardware fault. Run FurMark for 10 minutes. If it crashes or artifacts appear, RMA the card. That's the end of the line – I've only seen this twice in 14 years.