Fix DXGI_ERROR_DEVICE_REMOVED Game Crash

You're 40 minutes into a raid, framerate's fine, then the screen freezes, goes black for a second, and the game dies with "DXGI_ERROR_DEVICE_REMOVED" or code 0x887A0005. Sometimes it's on the main menu. Sometimes it's only after 1440p ultra. The crash pattern is the tell — if it hits under load, you've got a GPU problem. If it hits randomly even idle, you've got a driver problem.

This error isn't a Windows bug and it's not the game's fault, no matter what the dev forums say. DirectX is telling you the GPU stopped responding. The OS gave up waiting on it and told the app the device is gone.

What's actually happening

Windows runs something called TDR — Timeout Detection and Recovery. Every GPU call has a 2-second budget by default. If the GPU doesn't answer in that window, Windows resets the driver and every DirectX app on the machine gets DXGI_ERROR_DEVICE_REMOVED. The GPU is fine physically 99% of the time. What broke the 2-second window is the real question.

The culprit here is almost always one of four things: an unstable overclock (including factory OC cards), a bad driver install, overheating, or a game bug that hammers a specific feature like ray tracing. Nvidia 551.xx and 552.xx drivers had a nasty run of this on RTX 40-series cards in 2024. AMD's 24.5.1 had the same problem. If you're on those versions, that's your first clue.

The fix

Kill your overclock first. This is step one for a reason. GPU core, memory, and even CPU overclocks can trigger this. Reset the GPU to stock in MSI Afterburner. If your card is factory overclocked (most are), use the manufacturer's software or Afterburner to drop the core clock by -50 MHz and memory by -100 MHz. Undervolts count too — a too-aggressive curve causes this constantly on RTX 30 and 40 series. Test for an hour before moving on.
Clean install the driver with DDU. Don't use the Nvidia/AMD "clean install" checkbox. It doesn't clean anything. Boot into Safe Mode, run Display Driver Uninstaller, remove everything, reboot, then install the latest Game Ready / Adrenalin driver. If the newest driver is what broke you, roll back one version.
Check temps under load. Run HWiNFO64 alongside your game and watch GPU hotspot. If the hotspot is above 105°C on Nvidia or 110°C on AMD, you've got a thermal problem, not a software one. Repaste the card or fix your case airflow. A founders-edition 3080 with dried paste will absolutely throw this error at 1440p.
Turn off GPU acceleration in overlays. Discord overlay, Xbox Game Bar, Steam overlay, GeForce Experience overlay — pick one, disable the rest. Conflicting hooks cause device-removed crashes more often than people admit. This is the fastest test: disable all of them and play for two hours.
Raise the TDR delay. This isn't a fix, it's a workaround — but it stops the crash while you diagnose. Open Registry Editor as admin and go to:
```
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers
```
Create a DWORD (32-bit) called TdrDelay and set it to 10. Create another called TdrDdiDelay and set it to 10. Reboot. If the crashes go from "every 20 minutes" to "never," your GPU is taking 2-10 seconds to respond to something. That's still a problem, you just hid it.
Test the game with DX11 instead of DX12. Add -dx11 or -d3d11 to the Steam launch options, or find the equivalent setting in-game. If DX11 runs clean and DX12 crashes, it's a game or driver bug on the DX12 path. Report it and play on DX11 until it's patched. Don't waste hours fighting it.

If it still crashes

Run a clean Windows install test first — a spare SSD with a fresh Windows and only the GPU driver. If it still dies, you have a hardware problem.

Then check, in this order:

PSU. A failing or underpowered PSU causes transient voltage drops under load that reset the GPU. A 3080 wants 750W minimum from a decent unit, not a 600W no-name. If your PSU is 5+ years old, swap it.
The card itself. Run OCCT's VRAM test for 30 minutes. If it errors, the memory is degrading — RMA time.
PCIe slot and riser. If you're using a riser cable, remove it. Failing risers cause exactly this error, and it's the one people never check. Reseat the card too.
Game-specific patches. Some games ship broken DX12 implementations. Check the publisher's known-issues page — if it's a known bug, stop troubleshooting your own hardware.

Don't bother with "verify game files" or reinstalling the game first. It fixes this maybe 2% of the time. Go after the driver and overclock before anything else.

Drop a comment if you've got a weird one — the aftermarket 7900 XTX cards with bad vapor chambers are notorious for this, and no amount of driver rolling fixes those.