Fix GPU Driver Crashes: TDR, Black Screens, and Game Freezes

You're 40 minutes into a raid and the screen goes black. Fans spin up, then the whole thing recovers — or it doesn't, and you're staring at a reboot. Event Viewer says Event ID 4101, display driver nvlddmkm or amdkmdag stopped responding and has recovered. That's TDR — Timeout Detection and Recovery. Windows got tired of waiting for the GPU to respond and reset the driver.

The culprit is almost always one of four things: a bad driver install, a power delivery issue, an unstable overclock, or thermals. Work through this in order. Stop when the crashes stop.

Step 1 — The 30-second fix (try this first)

Nine times out of ten, the game itself is what's crashing — not the driver. Before you touch anything else:

1. Verify game files. Steam: right-click the game → Properties → Installed Files → Verify integrity. Epic: the three dots → Manage → Verify. A single corrupted shader cache file will crash you to desktop every time you hit the same spot in a level.
2. Close the overlay garbage. Discord overlay, GeForce Experience overlay, MSI Afterburner, RTSS, and Xbox Game Bar have all caused TDR crashes for me. Disable every overlay and retest.
3. Kill any overclock. If you're running a GPU overclock in Afterburner or a CPU overclock in BIOS, reset to stock. Unstable RAM XMP profiles cause the same 0x00000116 bugcheck, and people never suspect the RAM.

If it still crashes, move on.

Step 2 — The 5-minute fix (reset the driver stack)

Windows Update has been known to push a WHQL driver over your vendor driver, or leave half of the old one behind. The result is two driver versions fighting. Do this:

1. Win+X → Device Manager → Display adapters. Right-click the GPU → Properties → Driver tab. Note the driver version.
2. Right-click the GPU → Uninstall device. Check "Delete the driver software for this device."
3. Reboot. Windows will fall back to the Microsoft Basic Display Adapter. Screen will look awful. That's fine.
4. Install the current driver only — not the "Game Ready" bundle with GeForce Experience unless you actually use it. For NVIDIA, the Studio driver (yes, the creator one) is more stable in my experience. AMD's recommended WHQL driver, not "Optional."

Skip the "clean install" checkbox in the NVIDIA installer. It's superficial. If you want a real clean, use DDU — that's Step 3.

Step 3 — The 15-minute fix (DDU, registry, power)

When Steps 1 and 2 don't do it, you've got leftover driver files or a power delivery problem. Do these in order.

3a. DDU (Display Driver Uninstaller) — the real clean

Download DDU from Guru3D. Boot into Safe Mode (hold Shift while clicking Restart → Troubleshoot → Advanced → Startup Settings → Safe Mode). Run DDU, select your GPU vendor, click "Clean and restart." Then install the fresh driver while still disconnected from the internet so Windows Update can't sneak a driver in.

This fixes maybe 60% of stubborn TDR crashes. The other 40% are hardware or power.

3b. Check the TDR delay registry key

Default TDR delay is 2 seconds. Some games and some heavy compute workloads (Blender, Stable Diffusion) exceed that legitimately. You can raise it, but be aware: this masks the symptom. If you have a genuinely crashing GPU, a longer delay just means it hangs longer before it dies.

Path: HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers]
"TdrDelay"=dword:0000000a
"TdrDdiDelay"=dword:0000000a

Reboot. If crashes stop, great — but understand you've just told Windows to be more patient, not fixed the underlying hang.

3c. Power — the one nobody checks

A 3080 or 7900 XT draws 320-350W under load with transient spikes north of 500W. A tired 650W bronze PSU will trip exactly when a game hits a heavy scene. Symptoms: crashes start happening after the system's been running a while, or specifically during cutscenes, loading screens, or physics-heavy moments.

If your GPU is a 3-slot card and you're using the daisy-chain pigtail on the PCIe cable instead of two separate cables from the PSU, you've found your crash. Use separate cables. Every time.

Also: reseat the GPU. I've fixed more "dead" GPUs than I care to admit by just pulling the card, blasting the PCIe slot with compressed air, and putting it back firmly.

3d. Thermals and VRAM

Run HWiNFO64 in the background while gaming. Watch GPU hotspot temp (not the edge temp — the hotspot). If hotspot is over 105°C, you've got a thermal pad problem. Cards more than 2-3 years old often need a repaste and fresh thermal pads. Reference RX 5700 XT, RTX 3080 FE, and blower-style 2080 Ti are known offenders. This is a 30-minute job with a screwdriver and a tube of Kryonaut.

Quick diagnostic table

One more thing: if you've done all this and it still crashes, test the GPU in another machine before you buy a new one. And check the card's warranty — most are 3 years, and TDR crashes that survive a DDU in Safe Mode are usually a dying VRAM chip. That's an RMA, not a software problem.