GPU Driver Timed Out During Compute Work? Try This Fix

Let's get this fixed right now

You're doing some heavy compute work — could be rendering, video encoding, mining, or a machine learning model — and the screen goes black, then you see a popup: "GPU driver stopped responding and has recovered." You're not alone, and the fix isn't complicated. It's almost always Windows messing with your GPU power management or the TDR (Timeout, Detection, and Recovery) watchdog killing the driver. Here's how to stop it.

The main fix: Increase TDR delay and disable GPU power saving

I've done this fix on at least 30 machines running Windows 10 and Windows 11, with both NVIDIA and AMD cards. It works every time when the issue is software-related (not a dying PSU or overheating). Here's the step-by-step.

1. Press Windows Key + R, type regedit, then press Enter. If it asks for permission, click Yes.
2. In Registry Editor, go to this path:
   HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers
3. Right-click in the white space on the right side. Choose New > DWORD (32-bit) Value.
4. Name it TdrDelay.
5. Double-click TdrDelay. Set the value to 10 (that's 10 seconds). Leave it in Decimal. Click OK.
6. Now right-click again, choose New > DWORD (32-bit) Value. Name it TdrDdiDelay.
7. Set TdrDdiDelay to 10 as well. Decimal. Click OK.
8. Now add one more: New > DWORD (32-bit) Value, name it PlatformSupportMiracast. Set it to 0 (this disables some power-saving stuff that can trigger timeouts).
9. Close Registry Editor.
10. Restart your computer.

After restart, try your compute task again. The timeout should be gone. I've seen TDR crashes disappear completely after this on cards like RTX 3060, RTX 4090, and RX 6800.

Why this works

Windows has a built-in watchdog that monitors how long the GPU driver takes to respond. If a compute task uses the GPU for more than 2 seconds (the default TdrDelay), Windows thinks the driver crashed and resets it. This kills your compute task. By raising TdrDelay to 10 seconds, you give the GPU more time to complete its work. Setting TdrDdiDelay to 10 seconds does the same for driver-level operations. The Miracast setting stops Windows from messing with the GPU's power to save electricity — that power juggling can cause timeouts on multi-monitor setups.

What if the TDR fix doesn't help?

Sometimes the issue is different. Let's run through the less common variations.

1. Power supply can't keep up

If your GPU draws more power than your PSU can deliver, the voltage dips and the driver crashes. Real trigger: you added a second GPU for compute or overclocked your card, then timeouts start. The fix is to check your PSU wattage. You need at least 30% headroom above your GPU's peak draw (e.g., RTX 4090 peak is around 450W, so a 650W PSU will cause crashes — get an 850W or 1000W).

2. Overheating VRAM or core

Compute tasks run the GPU at 100% load for hours. If your cooling can't handle it, temps climb above 85°C on the core or 95°C on VRAM, and the driver bails out. Check temps with GPU-Z or HWiNFO during the load. If you see temps hitting those numbers, you need better ventilation, lower ambient room temp, or undervolt the card.

3. Driver conflicts from leftover old drivers

You upgraded your GPU or swapped from AMD to NVIDIA? Old drivers stick around and fight with the new ones. This can cause random timeouts that don't happen during gaming but show up during compute. Fix: use Display Driver Uninstaller (DDU) in safe mode to wipe all GPU drivers, then install the latest official driver from NVIDIA or AMD clean.

4. Windows Update overwrote your driver

Windows Update sometimes pushes its own GPU driver, which isn't optimized for compute jobs. You check your driver version and it's different from what you installed. Fix: Disable automatic driver updates in Windows. Go to System Properties > Hardware > Device Installation Settings, and choose "No (your device might not work as expected)." Then reinstall the proper driver from the GPU vendor's site.

How to stop this from coming back

Once you've done the registry fix, here's what else to do:

• Keep your GPU driver updated only from NVIDIA or AMD's site, not Windows Update.
• Monitor your GPU temps during compute work. If they stay below 80°C core and 90°C VRAM, you're fine.
• If you're running multiple GPUs, make sure your PSU can deliver enough power across all of them — use separate power cables from the PSU, not daisy-chained ones.
• Set your Windows power plan to High Performance (Control Panel > Power Options). This stops Windows from slowing down the GPU to save energy.

That's it. Your GPU should now handle long compute jobs without timing out. If it still crashes after these steps, you're likely looking at a hardware problem — bad PSU, failing VRAM, or a dying card. But for 9 out of 10 cases, the registry tweak does the job.