DXGI_ERROR_DEVICE_REMOVED means Direct3D lost its graphics adapter, while GetDeviceRemovedReason returns DEVICE_HUNG, DEVICE_RESET, DEVICE_REMOVED, DRIVER_INTERNAL_ERROR, or INVALID_CALL. Those API codes report adapter state, not proven cable or slot faults. Pair the reason code with single-game versus multi-game scope and recent changes to check stock clocks, drivers, and Event 4101 before opening the PC, and leave TdrDelay alone.
Contents
At a glance
- Class
- Tutorials
Provenance
- Oct 2, 2026Published
- Nov 13, 2026Last substantive update
A game that closes on a DXGI_ERROR_DEVICE_REMOVED box has lost its logical Direct3D graphics device. Most crash boxes and logs carry a second code — written as "with Reason", "DeviceRemovedReason", or a second 0x887A... hex number — and separating what that API return code actually measures from physical hardware guesses is what keeps you from tearing down a working PC.
In DirectX, DXGI represents your display subsystem (GPU, video memory, and outputs) through IDXGIAdapter, while the game renders through a logical Direct3D 11 or 12 device bound to it. When presenting a frame fails because that logical device was invalidated, the game calls ID3D11Device::GetDeviceRemovedReason (or its Direct3D 12 counterpart) to ask Windows why. Microsoft’s reference lists the possible return codes: DXGI_ERROR_DEVICE_HUNG, DXGI_ERROR_DEVICE_REMOVED, DXGI_ERROR_DEVICE_RESET, DXGI_ERROR_DRIVER_INTERNAL_ERROR, DXGI_ERROR_INVALID_CALL, and S_OK. Epic’s GPU crash guide for Unreal Engine shows how both appear together in a crash log: "with error DXGI_ERROR_DEVICE_REMOVED with Reason: DXGI_ERROR_DEVICE_HUNG".
Older guides call this "D3D device lost". That was a DirectX 9 term for window-focus transitions. Microsoft’s device-removed walkthrough explains that Direct3D 11 replaced it: the underlying IDXGIAdapter can still reset, switch adapters, reload its driver, or drop offline beneath a running game, and all of those surface through the DEVICE_REMOVED family of HRESULT codes.
Read your dialog
Find the second code (GetDeviceRemovedReason) in this table. Every code below is a user-mode Direct3D/DXGI API status, not a physical sensor on your PCIe slot or power supply cables. Even 0x887A0005 ("physically removed... or a driver upgrade") fires in software whenever Windows swaps or restarts a WDDM display driver, switches between integrated and discrete GPUs, or disables an adapter after a driver fault. The definitions come from Microsoft’s DXGI_ERROR reference and device-removed walkthrough.
| Returned reason code | Hex | Microsoft’s definition | What the API code actually tells a player (before touching hardware) |
|---|---|---|---|
DXGI_ERROR_DEVICE_HUNG | 0x887A0006 | “The application’s device failed due to badly formed commands sent by the application.” The walkthrough adds that the driver “stopped responding”. | A GPU workload timed out (TdrDelay) and Windows reset the driver. Points first at one game’s DX12/shader path, the driver version, or unstable GPU/CPU/RAM clocks — not a loose cable. |
DXGI_ERROR_DEVICE_REMOVED | 0x887A0005 | “The video card has been physically removed from the system, or a driver upgrade for the video card has occurred.” The walkthrough adds “turned off” or adapter switching. | Direct3D lost its IDXGIAdapter instance. Background Windows Update driver installs, laptop iGPU/dGPU handoffs, WDDM driver stops, PCIe link resets, and 12V power drops all produce the exact same 0x887A0005 code — check logs and scope before opening the PC. |
DXGI_ERROR_DEVICE_RESET | 0x887A0007 | “The device failed due to a badly formed command. This is a run-time issue.” | Another GPU context or runtime command triggered a shared adapter reset. Epic notes this often comes from a separate 3D app, browser, encoder, or overlay crashing in the background. |
DXGI_ERROR_DRIVER_INTERNAL_ERROR | 0x887A0020 | “The driver encountered a problem and was put into the device removed state.” | An internal fault inside the graphics driver. Clean-reinstall the latest vendor driver, or roll back if crashes started right after updating. |
DXGI_ERROR_INVALID_CALL | 0x887A0001 | “The application provided invalid parameter data.” | The game or rendering middleware passed invalid parameters to Direct3D. Patch the game, clear its shader cache, or switch API modes; hardware teardown cannot fix an invalid API call. |
Two limits apply to that table. Microsoft writes those strings for engine developers, so "the application" means the game binary rather than anything you clicked. And Epic cautions that GetDeviceRemovedReason codes "are not completely reliable" on their own because a driver stall can report HUNG on one run and REMOVED on the next. Treat the code as one of three inputs alongside scope and recent changes.
One rule holds across every code: Microsoft’s TDR overview notes that a DirectX 10 or later app can catch device removal, destroy and re-create its Direct3D device, and keep running. Most 3D games choose to show this dialog and exit instead.
One game or every game?
Scope narrows the fault faster than the hex code alone. Check three things before changing settings or touching the case:
- One game, or one game since its last patch, while other DirectX 12 games run clean at the same settings. Start with steps 1 to 3 below. That title’s rendering path or its pairing with your current driver is the primary suspect.
- Multiple DirectX games crash, or the Windows desktop flickers or goes black too. That points to shared software state (a broken driver install, background 3D app, or unstable GPU/RAM clocks) before physical hardware. Work steps 4 to 7 in order.
- It started right after a driver update, Windows Update, game patch, or GPU tuning tweak. Undo or test that exact change first.
Pair reason code, game scope, and recent change
Match all three signals — the GetDeviceRemovedReason code, single-game versus multi-game scope, and what changed most recently — to run the least invasive corroborating check first:
| Returned reason + scope | Recent change clue | Least invasive corroborating check (run first) |
|---|---|---|
DEVICE_HUNG (0x887A0006) or INVALID_CALL (0x887A0001) in one game | Game patch, DLC drop, or new graphics setting | Read the game’s crash log (Saved\Crashes in Unreal titles), test DirectX 11 mode or a 60 FPS cap, and check publisher patch notes (steps 1–3). |
DEVICE_RESET (0x887A0007) in one or more games | Discord/Steam/recorder overlay, hardware-accelerated browser, or second 3D app open | Fully restart Windows, close every background 3D/video app and overlay, and check System Event ID 4101 timestamps (step 1). |
DRIVER_INTERNAL_ERROR (0x887A0020) or any code across games | New GPU driver or cumulative Windows Update | Check Device Manager → Display adapters → Properties → Events for silent driver swaps, then run a clean DDU or AMD Cleanup Utility rollback to the last stable driver (step 2). |
DEVICE_HUNG (0x887A0006) across multiple games | GPU undervolt/OC, factory OC boost mode, PBO/Curve Optimizer, or XMP/EXPO | Query Event ID 4101 below, then return GPU clocks/voltage, CPU offsets, and memory profiles to stock defaults before retesting (step 4). |
DEVICE_REMOVED (0x887A0005) on a laptop or multi-GPU desktop | Battery/AC switch, MUX/Optimus toggle, or external monitor hot-plug | Pin the game executable to High performance in Settings → System → Display → Graphics so Windows stops switching IDXGIAdapter mid-session (step 5). |
Repeated DEVICE_REMOVED (0x887A0005) across multiple games at stock clocks | None (persists after clean driver + stock clocks) | Inspect System logs (Display 4101, nvlddmkm/amdwddmg, WHEA-Logger Event 17 PCIe bus errors) and Device Manager status (Code 43) before powering down to check PCIe/PSU seating or testing a known-good swap (step 7). |
Run the read-only Event Viewer check right after a crash to see whether Windows logged a WDDM timeout at that exact minute. Press the Windows key, type powershell, press Enter, and paste:
Get-WinEvent -FilterHashtable @{LogName='System'; ProviderName='Display'; Id=4101} -MaxEvents 5 | Format-List TimeCreated, Message
Event 4101 comes from Microsoft’s archived Event ID 4101 page under source Display: "Display driver %1 stopped responding and has successfully recovered." Finding Event 4101 at the crash timestamp confirms that WDDM detected a hung GPU workload and reset the driver — the OS half of the event whose game-side half is your DXGI dialog. Ignore that archived Vista/Windows 7 page’s old suggestion to lengthen the registry timeout; Microsoft’s current WDDM documentation overrides it, as Two fixes to skip explains.
If PowerShell prints "No events were found that match the specified selection criteria", Windows did not record a standard Display 4101 timeout for that exit. On Windows 11 25H2 build 26200.9457, verified on October 2, 2026, that command runs without administrator rights.
The fix order
The order is mine, cheapest first, with the vendor behind each step named. Test the game after each one.
1. Restart, and close other 3D apps
EA’s Battlefield 6 error-codes page gives one instruction for DXGI_ERROR_DEVICE_HUNG: fully restart your system. Epic explains why other programs matter. A GPU timeout “invalidates all currently running contexts”, so another program’s GPU crash takes your game down with it. Close browsers with hardware acceleration, video encoders, overlays and any second game before you relaunch.
If the crash never returns after a restart, you are done.
2. Update the game and the driver
Epic tells anyone reporting a GPU crash to use the latest driver and not to rely on Windows Update or even vendor tools to say it is current. Get it from the AMD, Intel or NVIDIA site. EA’s advice for AMD driver errors in Battlefield 6 is the same, with one addition: remove the old drivers first.
If the crash started with a new driver, roll it back instead. The clean removal, with AMD’s tools or DDU in Safe Mode, is step 1 of the GPU driver crash fix. On a Radeon card with DX12 crashes in Apex Legends or War Thunder, read that page’s September 2026 section first, which collects what AMD said about those two games.
Update the game too. INVALID_CALL, and a HUNG reason that only one game throws, both point at the game’s code by Microsoft’s definitions, and only a patch changes that.
3. Try the game’s DirectX 11 mode
AMD gave this workaround for one case: DX12 driver timeouts in Apex Legends and War Thunder on Adrenalin 26.8.1, as the GPU driver crash fix records. For other games it is my own suggestion. If the game offers DX11 beside DX12, switch for a session. A crash that stops in DX11 puts the fault in that game’s DX12 path or the driver’s handling of it, and you can keep playing while a patch arrives. The trade-offs between the two are in DX12 vs DX11.
4. Put everything back to stock
Intel’s DXGI_ERROR_DEVICE_HUNG article was written for one case, Prepar3D v4 on a laptop with Intel HD Graphics 620. After a driver update, its next step is to correct unstable overclocking by setting voltage and speed to the system’s specifications. Stock means all of it:
- any GPU overclock, and any undervolt, including one that ran clean for months
- a factory-overclocked card’s boost profile, if its software offers a stock or silent mode
- CPU overclocks, Precision Boost Overdrive and Curve Optimizer offsets
- the XMP or EXPO memory profile
I count an undervolt that passed a benchmark as unproven here, because the crash comes in one game’s scenes and not in the benchmark’s. If you tuned the card with one of our guides, the way back is in the undervolting guide, the RTX 5080 undervolt guide and the RX 9070 XT undervolt guide. Play at stock for a few days. If the crash stops, bring the tuning back one piece at a time.
The same Intel article then caps the game at 60 FPS, tests the memory with a tool such as MemTest, and reinstalls the game as the last step. The frame cap is a cheap test on any game that crashes in its busiest moments.
5. Laptops: pin the game to the discrete GPU
Microsoft’s list of ways a device gets removed includes a system that “changes from a power-saving graphics adapter to a performance graphics adapter”. That takes two GPUs in one PC, such as a laptop with integrated and discrete graphics. Microsoft’s windowed games page gives the path. Open Start > Settings > System > Display > Graphics, select the game under Custom options for apps, choose Options, set Graphics preference to High performance, and select Save. Restart the game.
6. Desktop timeouts and blue screens go to the TDR page
If the Event 4101 check above found timeouts outside games, or the PC blue-screens with VIDEO_TDR_FAILURE, the problem is no longer one game’s dialog. The GPU driver crash fix works that path from a clean driver install down, and this page does not repeat it.
7. REMOVED in every game: power and the card itself
Microsoft’s walkthrough describes DXGI_ERROR_DEVICE_REMOVED (0x887A0005) as an adapter "physically removed, turned off, or a driver upgrade has occurred", and explicitly adds that repeated device-removed errors "can also indicate a hardware failure or a bug in the graphics driver." That distinction matters: even repeated REMOVED crashes at stock clocks across multiple games leave competing software and hardware causes open — a WDDM driver bug on your GPU architecture, Windows Update silently overwriting your display driver mid-session, a shared engine or upscaler runtime fault across two Unreal/DX12 titles, PCIe link signaling errors (especially through riser cables), PSU transient drops, or a failing card.
Run three zero-teardown checks before reaching for a screwdriver:
- Device Manager → Display adapters → right-click your GPU → Properties → Events. Check whether Windows Update installed a different driver package near the crash time or left the adapter in Code 43 (driver reported a device failure).
- Event Viewer → Windows Logs → System at the crash minute. Look beyond
DisplayEvent4101fornvlddmkmoramdwddmgdriver errors andWHEA-LoggerEvent ID17PCIe root-port warnings, which point at PCIe slot/riser signaling rather than PSU power cables. - Engine crash logs (for example
%LOCALAPPDATA%\<Game>\Saved\Crashesin Unreal Engine titles) to verify whether two crashing games actually hit the same shader/plugin fault or drop at unrelated moments under 3D load.
Only when software logs are captured — or when crashes pair with black-screen fan spin-ups or a GPU disappearing from Device Manager — move to physical inspection. Shut down Windows, flip the PSU rear switch off, unplug AC power, touch bare case metal to ground yourself, and check the physical path: verify every 8-pin or 12V-2x6/12VHPWR connector is seated flush and latched at both the GPU and modular PSU ends, test without third-party cable extensions or vertical PCIe risers, and reseat the card in the primary PCIe slot. From there, use PSU failure symptoms for load-related power cutoffs, graphics card not detected if the adapter disappears from BIOS or Device Manager, and GPU artifacts if visual corruption precedes the crash. When clean drivers, stock clocks, and verified cables still leave 0x887A0005 repeating across unrelated engines, only a known-good GPU or PSU test (in another PC or with a spare unit) separates a driver/board quirk from a failing card or supply.
Two fixes to skip
The TdrDelay and TdrLevel registry edits
Older fix pages lead with a registry edit under HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\GraphicsDrivers. Microsoft’s TDR registry key page is written for driver developers, and it says “End users shouldn’t manipulate these registry keys.”
TdrDelay is the timeout itself, 2 seconds by default. Raising it gives a hung GPU more time before Windows resets it, and the clock, driver or power fault that caused the hang is still there. Epic’s Unreal Engine guide does set TdrDelay and TdrDdiDelay to 60 seconds, but that guide is for developers whose engine sends long workloads while they build a game, and Epic says the change “will not resolve all crashes.”
The TdrLevel edit is worse. In Microsoft’s table, TdrLevelOff (0) is “Detection disabled”. Windows stops looking for a hung GPU at all. The feature exists, in Microsoft’s words, “to prevent the entire system from becoming unresponsive”, so with it off, a hang that would have closed one game can freeze the whole PC instead.
Reinstalling DirectX
Microsoft’s DirectX support page says DirectX 11.3 and 12 ship with Windows 10 and later, that updates come through Windows Update, and that there is no stand-alone package. The DirectX End-User Runtime that fix pages link “does not change the version of DirectX”. It installs legacy libraries such as D3DX9 for older games. It cannot repair DirectX 12, and it does nothing for a device-removed error. Skip that download entirely.
Why the advice online disagrees
Four vendors explain this crash four ways.
- Microsoft blames DEVICE_HUNG on badly formed commands from the application, and adds that repeated removals can mean hardware or a driver bug.
- EA’s Battlefield 6 page says of DEVICE_HUNG, “This error happens when a background app is stuck.”
- Intel’s DEVICE_HUNG page, written for Prepar3D v4 on Intel HD Graphics 620, runs from a driver update to clocks, a frame cap and memory, and its steps are all on the player’s side.
- Epic’s list of GPU crash causes runs to content errors, engine bugs, driver bugs, overheating, running out of video memory and other apps.
They are each describing the slice they can see. Microsoft writes for developers, so the application is the obvious culprit. The other three write for people who cannot change the game’s code. Their steps are restarts and drivers first, then clocks. The fix order above follows the same split. The cheap player-side steps come first, a game update covers the developer’s side, and hardware comes last.
Game notes
Battlefield 6. EA’s error-codes page, updated August 11, 2026, says DirectX errors “can mean there’s an issue with your drivers”, asks you to check the game’s minimum requirements, and gives a full restart as the fix for DXGI_ERROR_DEVICE_HUNG.
Unreal Engine games. Epic’s guide shows the log line quoted at the top of this page and says the reason is a separate field from the error code. It is written for developers, but the reading carries over. A DEVICE_RESET reason in an Unreal game means, by Epic’s account, that another application probably crashed the GPU.
Apex Legends on Radeon. AMD’s statements about DX12 timeouts in Apex Legends and War Thunder are dated and linked in the GPU driver crash fix.
For Baldur’s Gate 3, FiveM, Arc Raiders, Wuthering Waves and Monster Hunter Wilds, I found no publisher support page that addresses this error as of October 2, 2026. For those, the scope test and the fix order above are the route.
Sources
- Microsoft Learn, DXGI_ERROR (Winerror.h) (page shows last updated March 15, 2021)
- Microsoft Learn, IDXGIAdapter interface (dxgi.h)
- Microsoft Learn, Handle device removed scenarios in Direct3D 11 (page shows last updated October 20, 2022)
- Microsoft Learn, ID3D11Device::GetDeviceRemovedReason (last updated February 22, 2024)
- Microsoft Learn, WDDM support for timeout detection and recovery
- Microsoft Learn, TDR registry keys (last updated November 7, 2025)
- Microsoft Learn archive, Event ID 4101
- Microsoft Support, How to install the latest version of DirectX
- Microsoft Download Center, DirectX End-User Runtime Web Installer
- Microsoft Support, Optimizations for windowed games in Windows 11
- Epic Games, Dealing with a GPU Crash When Using Unreal Engine
- EA Help, Troubleshoot error messages in Battlefield 6 (updated August 11, 2026)
- Intel Support, How to Resolve Error Message DXGI_ERROR_DEVICE_HUNG (Prepar3D v4; last reviewed May 19, 2022)
Frequently asked questions
What does DXGI_ERROR_DEVICE_REMOVED mean?
What does 0x887A0006 DXGI_ERROR_DEVICE_HUNG mean?
What does 0x887A0005 with DeviceRemovedReason 0x887A0006 mean?
Is DXGI_ERROR_DEVICE_HUNG a sign my graphics card is dying?
Does reinstalling DirectX fix DXGI_ERROR_DEVICE_REMOVED?
Should I raise TdrDelay to fix DXGI errors?
How do I fix DXGI_ERROR_DEVICE_HUNG in Battlefield 6?
Why does DXGI_ERROR_DEVICE_REMOVED happen in only one game?
Sources and corrections
- Last updated
- Methodology
- See our methodology for research and review standards. It draws on 13 cited sources, listed below, each checked against the original page on the date shown beside it. No first-party hardware fault injection stands behind this page; every definition, return code, and vendor step is quoted or paraphrased from the documentation pages below, opened on 2026-10-02 and rechecked on 2026-11-13 when separating IDXGIAdapter / GetDeviceRemovedReason API status codes from physical power or PCIe teardown checks. The Event 4101 command was run read-only on 2026-10-02 on Windows 11 Home 25H2, build 26200.9457, in Windows PowerShell 5.1 without administrator rights, to confirm its syntax, that the Display provider writes to the System log, and the message it prints when the log holds no such event. The triage order pairs reason code, scope, and recent changes from those sources.
- Sources
- Microsoft Learn: DXGI_ERROR (Winerror.h) (page shows last updated March 15, 2021) accessed 2026-11-13
- Microsoft Learn: IDXGIAdapter interface (dxgi.h) accessed 2026-11-13
- Microsoft Learn: Handle device removed scenarios in Direct3D 11 (page shows last updated October 20, 2022) accessed 2026-11-13
- Microsoft Learn: ID3D11Device::GetDeviceRemovedReason method (last updated February 22, 2024) accessed 2026-11-13
- Microsoft Learn: WDDM support for timeout detection and recovery accessed 2026-10-02
- Microsoft Learn: Testing and debugging TDR during driver development, TDR registry keys (last updated November 7, 2025) accessed 2026-10-02
- Microsoft Learn (archive): Event ID 4101, Display driver timeout detection and recovery accessed 2026-10-02
- Microsoft Support: How to install the latest version of DirectX accessed 2026-10-02
- Microsoft Download Center: DirectX End-User Runtime Web Installer accessed 2026-10-02
- Microsoft Support: Optimizations for windowed games in Windows 11 accessed 2026-10-02
- Epic Games: Dealing with a GPU Crash When Using Unreal Engine (Unreal Engine 5.8 documentation) accessed 2026-10-02
- EA Help: Troubleshoot error messages in Battlefield 6 (updated August 11, 2026) accessed 2026-10-02
- Intel Support: How to Resolve Error Message DXGI_ERROR_DEVICE_HUNG (Prepar3D v4 on Intel HD Graphics 620; last reviewed May 19, 2022) accessed 2026-10-02
- Corrections
- Spotted an error? Email a correction with the affected line and what you found. The draft includes this page’s address. Verified changes are recorded here. Contact details.