Game Crashed With DXGI_ERROR_DEVICE_REMOVED? Driver Reset, DirectX, VRAM, Power & Hardware Diagnosis Guide

| HK GPU Fix

Game Crashed With DXGI_ERROR_DEVICE_REMOVED? Driver Reset, DirectX, VRAM, Power & Hardware Diagnosis Guide

Game Crashed With DXGI_ERROR_DEVICE_REMOVED? Driver Reset, DirectX, VRAM, Power & Hardware Diagnosis Guide

Key Takeaways

DXGI_ERROR_DEVICE_REMOVED means Windows' graphics layer has declared your Direct3D device unusable, so the game cannot keep rendering and crashes. The cause is not always hardware: a driver update, a TDR reset, a laptop GPU switch, an unstable overclock, or a bad game command can all trigger it. Copy the full message, then work through the steps in order.

What DXGI_ERROR_DEVICE_REMOVED Actually Means

DXGI — the DirectX Graphics Infrastructure — is the layer inside Windows that connects a game, through DirectX, to the graphics hardware. When a game reports `DXGI_ERROR_DEVICE_REMOVED` (error code `0x887A0005`), DXGI is telling the application that the Direct3D device it was using has stopped being usable. That does not mean the card has been physically pulled out of the machine. It means the system now treats the device as gone, and the game must rebuild the device and its swap chain before it can draw another frame. A well-written game attempts that rebuild; a game that does not handle the situation simply crashes and shows the error.

This is why the same crash can look so confusing. The error message names a graphics device problem, yet the underlying cause can sit in the driver, the game, the power supply, the clock settings, or the hardware itself. The message alone does not name the cause. The words attached to it, the game involved, and the exact moment it happens are what narrow it down — and that is what this guide is built around.

Microsoft's own documentation for Direct3D 11 lists four situations in which a device can be lost even though the older "lost device" behaviour of DirectX 9 no longer applies. The virtualised device interface in Direct3D 11 lets several programs share one graphics card, but the device can still be removed when:

  1. The graphics driver is upgraded — including a background update the user never noticed.
  2. The system switches between a power-saving GPU and a high-performance GPU — the dual-GPU switching found on many laptops.
  3. The display device stops responding and is reset — that is, a TDR (Timeout Detection and Recovery) event fires.
  4. The graphics card is physically connected or removed.

Once the error is returned, the application must release the old device resources and create a new device. That is application-level work, not something a player can fix from inside the game. Understanding this point matters, because "restarting the game made it go away" is not the same as "the problem is solved". If the underlying trigger is still there, the error will come back — usually at the worst moment, in the middle of a demanding scene.

The Six Reason Codes: One Error Message, Six Different Meanings

The same `DXGI_ERROR_DEVICE_REMOVED` message can carry different underlying reasons, returned by `ID3D11Device::GetDeviceRemovedReason()`. Microsoft documents six possible return values, and they point in quite different directions. This table is the backbone of the diagnosis: do not treat every device-removed error as the same fault.

Reason codeWhat Microsoft says it meansWhat it usually points to for the player
`DXGI_ERROR_DEVICE_HUNG` (0x887A0006)The driver stopped responding because the application sent an invalid combination of graphics commandsTends to point at the game or application — the commands themselves were the problem. If it repeats, suspect a game update or a setting.
`DXGI_ERROR_DEVICE_REMOVED` (0x887A0005)The device was physically removed, turned off, or the driver was upgradedAn occasional occurrence can be normal; the game is expected to rebuild the device.
`DXGI_ERROR_DEVICE_RESET` (0x887A0007)The device failed because of a badly formed commandTends to point at the application sending a wrong command. Repeated occurrences point at the game itself.
`DXGI_ERROR_DRIVER_INTERNAL_ERROR` (0x887A0020)The driver hit an internal error and reset the deviceTends to point at the driver — a damaged driver, or a driver version with a bug.
`DXGI_ERROR_INVALID_CALL`The application passed an invalid parameterAn application bug.
`S_OK`The device was enabled, disabled, or reset, but did not fail (for example, WARP)Not necessarily a fault at all.

The key distinction to carry into the rest of this guide: HUNG and RESET point at the commands the game sent; DRIVER_INTERNAL_ERROR points at the driver itself; a plain REMOVED with no further reason code most often comes from a driver upgrade, a dual-GPU switch, or a TDR reset. The extra word in the error box — HUNG, RESET, DRIVER_INTERNAL_ERROR, or nothing at all — is the first clue. Copy the complete error message, not just "a DXGI error popped up". That one habit separates a useful diagnosis from guesswork.

TDR and Device Removed: The Same Event, Two Different Endings

TDR — Timeout Detection and Recovery — is a built-in Windows mechanism. When the GPU does not respond to a graphics command within its preset window (about two seconds), Windows assumes the GPU has hung and resets the display driver, so that the whole machine does not need to blue-screen. The same underlying event can end in two very different ways, and the site already covers one of them in depth.

The recovered ending is covered in #185. There, the TDR reset succeeds, the driver comes back, and the picture returns. The player sees the familiar "Display driver stopped responding and has recovered" message, and the game — or the desktop — keeps going.

This article covers the unrecovered ending. Here, the TDR reset happens, but the game does not survive the device rebuild. It receives the signal that the device is gone, cannot or does not know how to rebuild it, and crashes with `DXGI_ERROR_DEVICE_REMOVED`. The player sees an error box and a crash to desktop — not a "recovered" message.

So one underlying chain — the GPU hangs, Windows resets the driver — produces two outcomes: recovery (#185) and a crash with an error code (#365). The two articles share the TDR mechanism, but the ending and the diagnostic focus are different. If your screen goes black for a second and then comes back, read #185. If the game throws a DXGI error and closes, you are in this article's territory.

A related distinction is worth stating early, because it shapes every step that follows. A blue screen with `VIDEO_TDR_FAILURE` is a system-level failure: the whole machine stops. An error box followed by a crash to desktop is an application-level failure: the game ends, but Windows and the desktop keep running. The two are different in severity and in where to look first. A generic crash to desktop with no DXGI message at all is a different problem again, covered in #157 — this guide only handles cases where the literal `DXGI_ERROR_DEVICE_REMOVED` message appears.

What Can Cause It: The Causes, Ordered by How the Evidence Points

The causes below come from the overlapping consensus of several independent troubleshooting guides and Microsoft's own documentation. They are organised by the situation that points at each one, not as a flat list to try at random.

1. A damaged, buggy, or mismatched driver. This is the most common branch. A driver upgrade that went wrong midway, a damaged installation, or a particular version with a bug can all leave the GPU unable to recover after a hang. The evidence for this branch: the error started appearing after a driver update, or it disappears when you roll back to the earlier driver version.

2. An unstable overclock or undervolt. This covers the GPU core clock, the memory clock, and also system memory running with XMP or EXPO enabled. Overclocking can make the GPU stop responding under heavy load, which triggers a TDR; unstable system memory can slow the rendering pipeline and indirectly trigger a DirectX timeout. Several guides mention this consistently. Treat it as a direction to suspect and test — not as a certainty. Note also that some factory-overclocked cards ship with aggressive boost clocks, and there are reports that this alone can be enough on some cards; that is a report, not a settled fact, and it should be tested rather than assumed.

3. Unstable power, including transient power spikes. Modern GPUs can draw very large momentary power spikes. If the power supply's protection trips, the GPU can briefly lose power and disappear from the system bus, and DXGI then reports the device as removed. The tell-tale sign is that the error appears in heavy-load moments — explosions, large scene loads, sudden effects. The site's #151 already discusses the two-sided nature of the "Prefer Maximum Performance" power setting: keeping the card at high clocks and high voltage even at light load may itself create transient spikes, so the setting is a test variable with arguments on both sides, not a guaranteed fix. Keep that two-sided framing here.

4. Overheating — core, hotspot, or memory junction. The card's own protection can shut the GPU down at extreme temperatures to prevent damage, and that shutdown triggers the error. No universal temperature number is given in this guide, deliberately: there is no single standard that applies to every card. The correct approach is to watch the core temperature, the hotspot temperature, and the memory temperature together, and look for an abnormal climb at the moment the error strikes.

5. A problem in the game itself. The game may send invalid graphics commands (the HUNG and RESET reason codes), a game update may introduce the problem, or only one particular scene may trigger it. The evidence: only one game is affected, other games are fine, or the error began right after a game update.

6. Dual-GPU switching on a laptop. Switching between the power-saving GPU and the discrete GPU is one of the four situations Microsoft lists explicitly. The evidence: a laptop, and the error appearing at the moment the load switches.

7. Background factors. A driver updating in the background, recording or overlay software, and hardware-accelerated GPU scheduling (HAGS) are all mentioned in various guides as reported factors. The evidence for these is weaker, so they belong in this guide as items to disable one at a time and test — not as conclusions.

8. A physical connection problem. Poor contact on the card's gold fingers, a loose power cable, or card sag pulling on the PCIe slot can all produce an error that mimics a physical removal — because, from the system's point of view, that is what happened. The bar for this branch should be high: for example, the card disappearing from Device Manager at the same time, or the error only appearing after the case has been moved. Do not ask the reader to take the card out without a reason.

The Diagnostic Decision Tree

Work through these steps in order. Each step narrows the field; skipping ahead is how people replace a healthy card.

Step 0: Write Down the Error Message

Before anything else, copy the complete error box text, including any reason code — HUNG, RESET, DRIVER_INTERNAL_ERROR, or none. Note which game, which scene (loading screen, explosion, a specific level, or random), how often it happens, and what changed just before it first appeared: a driver update, a game update, an overclock, or a hardware change. Without this step, everything after it is guessing. A screenshot of the error box, kept on the phone, is the simplest way to keep the exact wording.

Step 1: Is It Only One Game?

If only one game shows the error, the balance of suspicion shifts toward that game: verify the game files, check the game's official known issues, and try another game as a control. If several games — or any heavy 3D load — produce it, continue to Step 2. Note that "only one game" narrows the range; it does not prove the card is healthy.

Step 2: Did It Start After a Driver Update?

If the error began after a driver update, roll back to the driver version that was installed before the update and test with it. If the error disappears after the rollback, the evidence points at the driver version. The rollback logic here is borrowed in one sentence from #210, which covers update regressions in full — do not turn this step into a second article about updates. If the error is not tied to an update, or it continues after a rollback, move on to Step 3.

Step 3: Check the System Logs

Open Event Viewer, go to Windows Logs, then System, and look around the time of the error for `nvlddmkm` Event ID 153 on NVIDIA cards — the "display driver stopped responding" warning — or for a LiveKernelEvent 141 entry in the application log. The method for finding these is already documented in #185 and #151; one sentence of reference is enough here, and there is no need to rewrite it as another TDR guide. If either entry is present, a TDR genuinely fired, and the chain is GPU hang, then driver reset. Reliability Monitor can also show whether a hardware error was recorded at the same time.

If there is no system-level record at all, the balance shifts toward the game layer, or toward an occasional driver upgrade — and that is a reason not to jump to a hardware conclusion.

Step 4: Cleanly Reinstall the Driver

Start with the clean-install option built into the driver installer itself: NVIDIA's custom install offers "Perform a clean installation", and AMD offers a Factory Reset. For most cases, that is enough to deal with a damaged driver installation.

Only use DDU (Display Driver Uninstaller) when a normal reinstall has not worked, when you have switched back and forth between several versions, or when you have changed graphics card brands. Download the target driver first, run DDU in Safe Mode with "Clean and restart", and then install the driver. DDU is a removal tool, not an installer, and it should only be downloaded from Wagnardsoft's official channel. DDU is not a cure-all and it is not something to run on every update — the built-in clean install is the right tool for routine updates. If the error continues after DDU, a damaged driver installation is excluded, and the investigation moves to Step 5.

Step 5: Return to Default Clocks and Voltages

Turn off every overclock and undervolt — GPU core, memory, and XMP or EXPO on the system memory — and return to the factory defaults. Then test the same scene that produced the error. If the error disappears, the evidence points at a stability problem, and the overclock should not be reapplied until a stable value has been found by testing. Remember that a factory-overclocked card's own boost clock can itself be aggressive; that remains a reported possibility, not a settled diagnosis.

Step 6: Power, Temperature, and Physical Checks

Watch for signs of a power spike at the moment the error occurs, and watch the core, hotspot, and memory temperatures together for an abnormal climb. Check that the power cables are firmly seated. On a laptop, check whether the error lines up with a switch between the two GPUs. If this step finds nothing, do not force a power-supply verdict — a suspicion about power needs further checking, and one error is not enough to condemn a power supply.

Step 7: The Stopping Point

Stop testing under heavy load and record the results for inspection if any of the following apply:

Also stop if the same error keeps appearing frequently after a clean driver reinstall and a return to default clocks. At that point, further random setting changes will not help, and the card should be inspected with the test record in hand: the full error message, which game and scene, the steps already tried and their results, and the driver version.

Boundary Cases: When the Conclusion Changes

The error fires at the exact moment a background driver update runs. This is one of the situations Microsoft lists as normal: upgrading the driver itself invalidates the device once. Restart the game and, if the error does not come back, there is no fault to chase.

The error only appears on loading screens or at light load. The site's #151 covers this territory: at light load, high clocks held by "Prefer Maximum Performance" may create transient spikes, while the power-state switch itself is also a suspect. The correct handling is a controlled comparison — try "Optimal Power" for a few days, then "Prefer Maximum Performance" for a few days, and record how often the error appears under each. Do not pick one side as the answer.

A laptop with two GPUs. The error may have nothing to do with the discrete card and everything to do with the switching logic. Assigning the game to the high-performance GPU explicitly is one of the test items.

The same error code, but a different reason code. HUNG means the game's commands are suspect; DRIVER_INTERNAL_ERROR means the driver is suspect; no reason code at all most often means an upgrade, a switch, or a TDR. The article has to teach the reader to separate these three, not lump them together.

A blue screen versus an error box. A blue screen is a system-level failure and the whole machine stops; an error box is the game ending while Windows and the desktop keep running. They differ in severity and in where to look first. The FAQ of #157 already draws this distinction; one sentence of reference is enough.

Black screen versus an error box. Articles #357 and #358 cover a game going black or freezing on a single solid colour, where the picture fails but no DXGI error appears; this guide covers the opposite case, where the game throws a DXGI error code and closes.

Common Misconceptions

"A DXGI error means the card is dead and must be replaced." Wrong. Microsoft explicitly says that a driver upgrade and a dual-GPU switch — both normal events — can trigger it, and several independent guides agree that software and settings conflicts are a frequent cause. One error cannot condemn a card.

"Seeing 'device removed' means it is a hardware problem." Wrong. The FAQ of #157 already says this: the failure can come from a driver reset, from VRAM exhaustion, or from the game itself. The message has to be read together with the conditions and the system logs.

"Editing the registry to change the TDR timeout values is a fix." This is not taught in this guide, and the steps are deliberately not given. The misconception is already addressed in #157: changing those values only hides the symptom, turning a crash into a freeze, and the long-term risk is higher. For a device-removed error, lengthening the TDR wait is rarely effective anyway, because the problem is not that the wait was too short — the system believes the hardware is already gone. The correct repair is to find out why the card stalls in the first place.

"Reinstalling the game is the universal first step." Wrong, and #157 already says so. Reinstalling takes the longest and only fixes the damaged-files branch. Verifying the game files handles almost all file problems already.

"Normal temperatures mean the hardware is fine." Wrong, and #151 already says so. The GPU core temperature sensor does not reflect the memory or voltage-regulator temperatures, and overheating in either of those can cause the same error.

"Prefer Maximum Performance is always more stable." Wrong, and #151 covers both sides of this. There are counter-examples, and the setting should be treated as a comparison test, not a rule.

"Some guide online gives a precise-sounding percentage and says almost all cases are software, so the hardware can be ignored." Wrong. That percentage figure appears in at least one guide with no data source behind it, and it must not be quoted. The honest statement is that many cases sit at the software or settings layer, but each branch has to be excluded by following the decision tree — no percentage can replace a diagnosis.

"DDU is a cure-all, and should be used on every update." Wrong. DDU is for damaged drivers, repeated version switching, and brand changes. Routine updates are handled by the clean-install option built into the driver installer.

FAQ

Does one DXGI_ERROR_DEVICE_REMOVED crash mean my card is dying?

No. A single occurrence, especially one that lines up with a driver update in the background or a laptop GPU switch, can be entirely normal. What matters is the pattern: how often it repeats, whether it is tied to one game or to every 3D load, and what the system logs show. A card is only suspected after the decision tree has excluded the driver, the clocks, the game, and the power checks — and even then, the pattern is evidence, not a verdict.

What should I copy from the error box?

The complete text, including the reason code if one is shown — HUNG, RESET, DRIVER_INTERNAL_ERROR, or the plain code on its own. Also note the game, the scene, and what changed just before the error first appeared. That combination is what lets anyone reading the report tell the three main branches apart.

Why does restarting the game fix it, if the problem is real?

Because the game rebuilds a fresh device when it starts. If the trigger was a one-off event — a background driver update, a single bad scene — the new device works fine. If the trigger is still there, the error will return, usually in the same kind of scene. A restart that "fixes" it once is not proof the cause is gone.

My screen went black for a second and then came back, with no error. Is that the same thing?

No — that is the recovered ending, covered in #185. The driver reset and came back. This article covers the ending where the game does not survive the reset and crashes with the DXGI error instead.

The error only happens in one game. Does that mean the card is fine?

It narrows the range toward that game, but it does not prove the card is healthy. A game can be the only one that sends the particular command pattern, or the only one that stresses the card in the particular way, that exposes a wider instability. Use the one-game result as a direction, then verify the game files and test the other branches before concluding anything.

Should I change the TDR timeout values in the registry to stop the crash?

No. This guide does not give those steps, and the site's #157 explains why: changing the registry only hides the symptom and can turn a crash into a freeze. Find the reason the card stalls instead.

I rolled back the driver and the error went away. Can I stay on the old driver?

The rollback is a diagnostic test, and a disappearing error after a rollback is useful evidence pointing at the driver version. Whether to stay on that version is a separate decision — older drivers can miss fixes and security updates — and it should be made after checking whether a newer driver version later resolves the problem, rather than treating the rollback itself as the permanent repair.

When should I stop testing and bring the card in for inspection?

When the error comes with a blue screen, when the card disappears from Device Manager, when the machine boots with no picture, when there is a burning smell or an abnormal sound, or when the error keeps appearing frequently after a clean driver reinstall and a return to default clocks. Bring the test record with you: the full error message, the game and scene, the steps tried and their results, and the driver version.

When to Bring the Card In

If the decision tree ends without an answer — the driver is cleanly installed, the clocks are at factory defaults, the error still appears across several games, and the logs show repeated TDR events — the next step is a bench inspection rather than more setting changes. Bring the error screenshot, the list of steps already tried and what each one showed, and the driver version. That record is what turns a vague "it crashes" into a diagnosis that can actually be worked on.

HK GPU Fix is at Room 1209, Dali Plaza, Shek Mun, Shatin (石門達利廣場 1209 室). Phone: 2186 8010. WhatsApp: 6558 6806. Inspection is charged within the published three-tier range of HK$400–2,800, with the exact figure quoted after inspection according to the actual condition found.

Related articles

Related services

If you need VRAM upgrade / repair, see these service pages: