If you have ever been mid-match, suddenly greeted by a desktop crash bearing the text “DXGI_ERROR_DEVICE_HUNG,” you already know how maddening this error can be. DXGI_DEVICE_HUNG is a DirectX Graphics Infrastructure return code (0x887A0006) that fires when the GPU driver determines the graphics device has stopped responding to commands and cannot recover on its own. Unlike a clean in-game crash, this error severs communication between the engine and the GPU at the driver level, which is why the game window typically vanishes or freezes before Windows displays the error dialog.
- Likely Causes
- Fix 1: Disable all overlay software
- Fix 2: Delete and rebuild the shader cache
- Fix 4: Remove GPU overclocks and test at stock
- Fix 5: Change Windows power plan to High Performance
- Fix 6: Reduce VRAM pressure with texture and resolution settings
- Fix 7: Increase the TDR delay via registry
- If Nothing Works
- FAQ
Engines built on Unreal Engine 4 and 5 are the most frequent reporters of this error. You will see variants like “Fatal error! The reason: DXGI_ERROR_DEVICE_HUNG found at D3D11CreateDeviceAndSwapChain” or “A device attached to the system is not functioning” in titles such as Call of Duty: Warzone, Fortnite, Valorant (which uses a modified Unreal branch), and countless other UE-based games. Unity-based titles like Escape from Tarkov and Rust produce similar strings. Custom-engine games on Windows 10 and 11 also throw DXGI_DEVICE_HUNG when GPU timeout detection and recovery (TDR) fails to reset the adapter.
Likely Causes
| Cause | How to confirm | Fix difficulty |
|---|---|---|
| GPU overclock (core, memory, or power limit) exceeding stable thresholds | Check MSI Afterburner, AMD Adrenalin, or NVIDIA Control Panel for active offsets; test at stock clocks | Easy |
| Corrupted or incompatible graphics driver | Event Viewer shows Display driver nvlddmkm or amdkmdag stopped responding; error resolves after clean driver reinstall | Easy |
| Overlay software injecting into the D3D device | Error disappears with all overlays disabled (Discord, Steam, Xbox Game Bar, Rivatuner) | Easy |
| VRAM exhaustion or memory instability at high resolutions | Reproduces only above 1440p or with ultra textures; GPU-Z shows VRAM usage at 95%+ | Moderate |
| Insufficient or unstable power delivery from PSU | OCCT or HWiNFO shows 12V rail dropping below 11.4V under load | Moderate |
| Windows power plan set to a balanced or aggressive sleep profile | Error correlates with CPU package C-states or PCIe link power management | Moderate |
| Corrupted game files or shader cache | Error clears after deleting shader cache and verifying file integrity | Easy |
| Dying GPU hardware (VRAM errors, degraded power phases) | Multiple stress tools report artifacts or driver resets within minutes | Hard |
Fix 1: Disable all overlay software
How to check
Open the Discord overlay setting (Settings > Game Overlay > toggle off “Enable in-game overlay”). In Steam, right-click the game > Properties > General > uncheck “Enable the Steam Overlay while in-game.” On Windows 11, go to Settings > Gaming > Xbox Game Bar and turn it off. If you use MSI Afterburner with RivaTuner Statistics Server, check whether the on-screen display is active.
What to do
Disable every overlay source at once, launch the game, and play for at least twenty minutes to see if DXGI_ERROR_DEVICE_HUNG recurs. Re-enable them one at a time to identify the culprit. Some users report that Discord’s hardware-accelerated overlay interacts badly with Unreal Engine’s swapchain management.
How to undo it
Simply toggle each overlay setting back to its original state. No files are modified.
Fix 2: Delete and rebuild the shader cache
How to check
Navigate to %LOCALAPPDATA%\D3DSCache and check the folder size. For NVIDIA users, also inspect the DXCache folder at C:\Users\[username]\AppData\Local\NVIDIA\DXCache. If these caches exceed several gigabytes or were populated with a different driver version, they can cause submission errors during pipeline state creation.
What to do
Close the game. Open File Explorer, paste %LOCALAPPDATA%\D3DSCache into the address bar, and delete all contents. Do the same for the NVIDIA DXCache or AMD equivalent at %LOCALAPPDATA%\AMD\DxCache. Relaunch the game; it will recompile shaders, which may cause a longer first load and brief stuttering.
How to undo it
No undo needed. The cache regenerates automatically on the next run.
How to check
Open the NVIDIA Control Panel and check if any 3D settings overrides are active. On AMD, open Adrenalin > Graphics and look for custom profiles. The most reliable check is Event Viewer > Windows Logs > System, filtering for “Display” source. Look for entries reading “Display driver nvlddmkm stopped responding and has successfully recovered” or the amdkmdag equivalent.
What to do
Press Win+X, open Device Manager, expand Display adapters, right-click your GPU, and select “Uninstall device” with the “Attempt to remove the driver for this device” checkbox ticked. Reboot. Windows will install a generic driver, which will then let the Windows Update or your GPU vendor’s installer push a fresh version. For the cleanest result, use Display Driver Uninstaller (DDU) in Safe Mode, select “Clean and restart,” then install the latest stable driver from the vendor’s site rather than the optional branch.
How to undo it
There is no meaningful undo for a driver uninstall. The system will automatically install a fallback driver on reboot.
Fix 4: Remove GPU overclocks and test at stock
How to check
Open MSI Afterburner, AMD Adrenalin, or EVGA Precision X1. Note the core clock offset, memory clock offset, and any power limit increase. Even factory-overclocked cards benefit from manual offsets users may have set months ago and forgotten about.
What to do
Reset all clock offsets to zero. If your power limit slider was raised above 100%, set it back to the default. Save the profile, apply, and test the game for at least an hour. If the error vanishes, your GPU’s silicon cannot sustain the previous frequencies under the specific load patterns of this engine.
How to undo it
Re-enter your previous offset values in the overclocking tool and click Apply.
Fix 5: Change Windows power plan to High Performance
How to check
Open Control Panel > Hardware and Sound > Power Options. If “Balanced” is selected, the PCIe link power management and processor idle states may cause the GPU to enter a low-power state mid-frame, triggering a TDR failure that DXGI surfaces as a device hung error.
What to do
Select “High Performance.” Click “Change plan settings” > “Change advanced power settings.” Expand PCI Express > Link State Power Management and set both On Battery and Plugged In to “Off.” Expand Processor power management > Processor performance core parking min cores and set it to 100%. Apply and reboot.
How to undo it
Return to the Balanced plan and revert each advanced setting to its default value (typically “Moderate” or “10%”).
Fix 6: Reduce VRAM pressure with texture and resolution settings
How to check
While the game is running, alt-tab and open Task Manager > Performance > GPU. Watch dedicated GPU memory usage. If it approaches or exceeds your VRAM capacity (e.g., 7900 MB on an 8 GB card), the driver may be thrashing to system RAM and eventually timing out.
What to do
Lower the texture quality one step from Ultra to High. If you play at 4K, try 1440p or enable DLSS/FSR with texture quality set to Medium. In Unreal Engine games, add -dxgi=11 to the launch options in Steam or the Epic Games Launcher to force a specific swapchain mode that sometimes behaves more predictably under memory pressure.
How to undo it
Revert in-game settings to your previous values and remove the launch option string from the launcher properties.
Fix 7: Increase the TDR delay via registry
How to check
Open the registry editor and navigate to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers. Look for the TdrDelay and TdrDdiDelay DWORD values. If they are absent, the default is 2 seconds, which may be too aggressive for a GPU briefly saturating during shader compilation.
What to do
Create a new DWORD (32-bit) value named TdrDelay and set it to 8. Create another named TdrDdiDelay set to 8. Reboot. This gives the driver eight seconds to respond before Windows declares a hang, allowing the engine to survive longer hitch moments without a hard device lost event.
How to undo it
Delete both registry keys you created and reboot, restoring the system defaults.
If Nothing Works
If you have cycled through every fix above and DXGI_ERROR_DEVICE_HUNG still appears with predictable regularity, the problem has likely shifted from software to hardware. Run OCCT’s VRAM test for at least one full cycle and the GPU 3D adaptive test simultaneously while monitoring the 12V rail with HWiNFO64. If OCCT reports memory errors or the rail voltage sags below 11.4V, your power supply may be failing under transient loads that modern GPUs demand in 2026. Alternatively, a GPU exhibiting consistent driver resets across multiple titles and driver versions is a strong indicator of a degrading power phase or solder fatigue on the VRAM modules. In that case, contact the manufacturer’s warranty service before the window closes, and document your error codes with screenshots.
FAQ
Is DXGI_DEVICE_HUNG a sign my GPU is dying?
Not necessarily. Most reports resolve after a clean driver install or by reverting an overclock. Treat hardware failure as a last diagnosis after you have eliminated software, power, and thermal causes. However, if the error appears across multiple unrelated games and persists after a DDU reinstall, hardware is the probable culprit.
Does this error damage my hardware?
The error itself is a notification that Windows reset or gave up on the GPU adapter. No data is written to the hardware that would cause degradation. The underlying cause (voltage sag, thermal spike) can stress components over time, but the DXGI device hung event is benign.
Why does the error only happen in one game?
Each engine builds different command submission patterns, pipeline states, and swapchain configurations. A game using Unreal Engine 5 with Nanite and Lumen places a specific microsecond-level load on the GPU’s command processor that other titles do not replicate. This can expose marginal stability that other games never trigger.
Will upgrading to Windows 11 24H2 help?
Some users report that the updated DXGI scheduler in recent Windows 11 builds handles GPU resets more gracefully, reducing the frequency of the error dialog even when a TDR event still occurs. This is not a universal fix and should be tested alongside the steps above rather than treated as a standalone solution.
