Fix 'GPU Device Removed' Crash in Halo Infinite on Windows 10/11

You're mid-firefight on Recharge, the screen stutters for half a second, then Halo Infinite dumps you to the desktop with something like 0x887A0005 or DXGI_ERROR_DEVICE_REMOVED in the event log. The GPU didn't die. Windows yanked it out from under the game because something went sideways for longer than the driver could hide.

What's actually happening here is Windows' Timeout Detection and Recovery (TDR) watchdog. If the GPU doesn't respond within 2 seconds — the default — Windows resets the adapter. DirectX surfaces that to the game as "device removed." Halo Infinite is unusually sensitive to this because it holds longer command buffers than most shooters and leans hard on async compute on RDNA and Ada cards. In almost every case I've seen, it's one of three things: TDR losing a race it shouldn't, a driver version with a known bug, or an overclock that's been quietly unstable for months.

Work through these in order. Don't start with the exotic stuff.

Cause 1: TDR is too aggressive for Halo Infinite's spikes

This is the big one. On Windows 10 21H2 and Windows 11 22H2 and newer, the default TDR delay is 2 seconds and the recovery limit is 1. Halo Infinite's shader compilation and level streaming can stall the GPU for slightly longer than that on a 1440p or 4K setup, especially the first time you spawn into a new map. Windows panics, resets the driver, and the game dies.

The real fix is to raise the TDR delay. This doesn't hide a broken GPU — it gives the driver enough headroom to finish the buffer it's working on. People have been doing this since Battlefield 3.

1. Press Win+R, type regedit, and hit Enter.
2. Go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers
3. Right-click in the right pane, create a new DWORD (32-bit) value named TdrDelay and set it to 10 (decimal).
4. Create another DWORD named TdrDdiDelay and set it to 10 as well.
5. Reboot. A reboot is required — the value is read at boot.

If you're on a laptop with switchable graphics, also check that Halo Infinite is running on the discrete GPU and not Intel UHD. Windows Graphics Settings → find HaloInfinite.exe → set to High performance. Running the game on an iGPU that shares memory with the CPU is a fast track to device-removed crashes.

Don't set TdrDelay to 60 or 120 "to be safe." If your GPU is genuinely hanging for a minute, you want the reset, not a frozen machine. 8–10 seconds is the sweet spot.

Cause 2: A driver version with a known Halo Infinite bug

Both NVIDIA and AMD have shipped drivers that crash Halo Infinite specifically. NVIDIA's 526.x branch had a well-documented device-removed issue with the game on 30-series cards. AMD's Adrenalin 22.11.2 through 23.2.x had intermittent TDRs on RX 6000 series under the campaign's Forerunner structures. If your driver is from either of those windows, no amount of registry tweaking will save you — the driver itself is the problem.

The fix is a clean reinstall, not an over-the-top one. Over-the-top installs leave stale kernel-mode DLLs behind and that's how you end up chasing ghosts.

1. Download DDU (Display Driver Uninstaller) from Wagnardsoft. Yes, it's still the right tool.
2. Boot into Safe Mode. DDU can do this for you from the Options menu.
3. Run DDU, select your vendor (NVIDIA or AMD), and choose Clean and restart.
4. Once Windows boots, grab the latest WHQL driver directly from the vendor's site. Don't let Windows Update install a generic one — it'll pick something old.
5. Install, reboot, then launch Halo Infinite.

On NVIDIA, go one step further: disable the GeForce Experience overlay for Halo Infinite. The overlay's frame capture hooks into the same present chain that's already fragile. I've seen it turn a stable 3-hour session into a crash every 20 minutes on a 3080. In GeForce Experience → Settings → General, turn off In-Game Overlay entirely while you're troubleshooting.

Also worth doing: turn off Nvidia Reflex temporarily. It can interact badly with the game's frame pacing on some driver versions. You can re-enable it once you've confirmed stability.

Cause 3: An unstable GPU overclock or undervolt

Here's the one people argue about. Your overclock that's been rock solid in Cyberpunk and 3DMark for a year can still fail in Halo Infinite. What's happening is that Infinite's shader compiler hits the GPU in a pattern that other games don't — short, sharp bursts of full utilization with rapid clock changes. A memory offset that's stable under sustained load can crash under that pattern.

Don't argue with me on this, just test it. Reset MSI Afterburner to stock (the little circular arrow), or if you're on AMD, reset the Adrenalin tuning profile to Default. Play for an hour. If the crashes stop, your OC was the cause and you now know your real stable ceiling.

Common culprits, in order of likelihood:

• VRAM offset too high — crashes within minutes of loading a new map. Back it off by 200 MHz.
• Core undervolt too aggressive — crashes during shader compilation on first boot. Raise the voltage floor by 25mV.
• Power limit too low — crashes when particles fill the screen on Behemoth or Live Fire. Raise the power limit.
• Factory OC card (e.g., Strix, Suprim) on a weak PSU — crashes under transient spikes. This is the one people miss. A 3080 Strix can pull 450W transients. A 650W PSU that's fine in every other game can brown-out in Halo Infinite's loading screens.

If you're on a factory-OC card and nothing else has worked, drop the power limit to 90% in Afterburner and play an hour. If the crash goes away, your PSU is the weak link, not the GPU.

Quick reference

One last thing: check Event Viewer under Windows Logs → System right after a crash. You'll see a Display entry with event ID 4101 — "nvlddmkm stopped responding and has successfully recovered" or the AMD equivalent. That's the confirmation that TDR fired. If you see a different error instead, like a WHEA-Logger entry, stop what you're doing and check your hardware — that's a real fault, not a Halo quirk.