Fix DXGI_ERROR_DEVICE_REMOVED (0x887A0005)

DXGI_ERROR_DEVICE_REMOVED (0x887A0005) means Windows completely lost contact with your GPU mid-game and treated it as if it were unplugged from the PCIe bus — so the game had no device left to render on and crashed. It is more severe than the timeout-based DXGI_ERROR_DEVICE_HUNG (0x887A0006): a corrupt driver, an unstable overclock, memory instability, or a power/seating fault yanked the graphics device off the system entirely. The fixes below are ordered from most to least common cause.

The GPU dropped off the bus — Windows saw it as removed. The cause is almost always driver, overclock, memory, or power.

What the error actually is

The full crash usually reads something like DXGI_ERROR_DEVICE_REMOVED or, from a Direct3D 12 title, DirectX function "GetDeviceRemovedReason" failed with DXGI_ERROR_DEVICE_REMOVED. The hex code is 0x887A0005.

DirectX exposes a helper, GetDeviceRemovedReason(), that returns why the device left. The sub-reason it reports narrows the cause:

Reason code	Hex	What it points to
DXGI_ERROR_DEVICE_HUNG	0x887A0006	GPU missed the ~2s TDR timeout — slow, not gone (see the DEVICE_HUNG guide)
DXGI_ERROR_DEVICE_REMOVED	0x887A0005	Full comms loss — driver crash, power, or seating
DXGI_ERROR_DEVICE_RESET	0x887A0007	A badly formed command reset the device — usually a game/driver bug
DXGI_ERROR_DRIVER_INTERNAL_ERROR	0x887A0020	The driver itself faulted — reinstall it clean

If the game logs a reason, note it: a DEVICE_HUNG reason means the fixes on our DXGI_ERROR_DEVICE_HUNG page apply too, since the two crashes overlap heavily.

Fix 1 — Clean-reinstall the GPU driver with DDU

A corrupt or half-updated driver is the single most common cause of DEVICE_REMOVED. A normal “update” installs over the broken files; a clean reinstall removes them first.

1. Download the latest driver from NVIDIA or AMD, plus Display Driver Uninstaller (DDU).
2. Boot into Safe Mode, run DDU, and choose Clean and restart.
3. Install the freshly downloaded driver. For NVIDIA, a Custom → Clean install is enough if you would rather skip Safe Mode.

This alone resolves the majority of cases. If it comes back, the cause is not the driver — move on.

Fix 2 — Disable XMP / EXPO and test RAM at JEDEC

This is the fix most guides miss. Your RAM’s XMP (Intel) or EXPO (AMD) profile runs memory above its guaranteed speed, and an unstable profile makes the CPU fail to feed the GPU cleanly — which drops the graphics device off the bus and surfaces as DEVICE_REMOVED, even though nothing is wrong with the GPU.

1. Enter BIOS (usually Del or F2 at boot).
2. Set the memory profile to Disabled (base JEDEC speed, e.g. 4800 MT/s).
3. Save, reboot, and play. If the crashes stop, the profile was the cause.

You do not have to give up XMP forever — re-enable it at a lower frequency or with looser timings once you have confirmed it was the trigger. If crashes continue at JEDEC, run MemTest86 overnight to rule out a genuinely faulty stick.

Fix 3 — Reset any GPU or CPU overclock

A factory or manual overclock that is stable in a benchmark can still fail under a real game’s mixed load and drop the device.

• MSI Afterburner / vendor tool: click Reset to return the GPU to stock clocks and voltage. See the MSI Afterburner setup guide for where the reset is.
• Factory-OC cards: even out-of-the-box “OC edition” cards can be unstable — test at a small negative core offset (−50 to −100 MHz).
• CPU: disable any manual CPU overclock or aggressive Precision Boost / MCE setting in BIOS.

An undervolt often fixes DEVICE_REMOVED where a downclock does not, because it lowers the voltage droop that causes the crash while keeping performance.

Fix 4 — Check GPU power and seating

DEVICE_REMOVED is a physical disconnect from Windows’ point of view, so bad power delivery is a real cause on desktops:

• Use separate PCIe power cables for each connector on high-draw cards — do not daisy-chain a single cable across two 8-pins.
• Reseat the GPU firmly in the top PCIe slot and reseat the power connectors at both ends.
• Confirm your PSU has enough headroom (a 4070-class card wants ~650 W; a 4080/4090-class card ~850 W).
• On a laptop, always play plugged in — battery power starves the dGPU and triggers this crash.

Fix 5 — TdrDelay registry workaround (last)

Raising the Timeout Detection and Recovery window helps a DEVICE_HUNG crash far more than a true DEVICE_REMOVED, so try this only after Fixes 1–4.

1. Open Registry Editor, go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers.
2. Create a QWORD (64-bit) value named TdrDelay, set to 8 (decimal).
3. Reboot.

This buys the GPU longer before Windows declares it lost. It is a workaround — if it “fixes” the crash, an unstable OC or driver is still the real cause underneath.

Still crashing?

Work down the list — driver, memory, overclock, power — before suspecting the GPU itself. If DEVICE_REMOVED persists at stock clocks, JEDEC RAM, a clean driver, and confirmed power, test the card in another system to isolate a dying GPU. For the related timeout crash, see how to fix DXGI_ERROR_DEVICE_HUNG; for the VRAM-exhaustion crash, see out of video memory. Keeping drivers clean and your system stable also smooths frame pacing and input lag — the same stability work pays off in-game.