VIDEO_TDR_FAILURE: Fixing GPU Timeout Blue Screens

VIDEO_TDR_FAILURE is bug check 0x00000116. The GPU stopped responding, Windows told the graphics driver to reset itself, and the reset failed. The file named on the blue screen identifies the vendor, and that name is where troubleshooting starts.

Nearly all of these come down to four things: a bad driver version, heat, an overclock, or power delivery. Three of the four cost nothing to test.

What TDR is, and why a failed reset stops Windows

Timeout Detection and Recovery exists so a busy graphics card does not take the desktop down with it. Microsoft’s driver documentation describes it plainly: TDR detects when the graphics card takes longer than expected to complete an operation and then resets the graphics card to keep the system responsive. The default timeout period in Windows is two seconds.

When recovery works you get a flicker and a notification saying the display driver stopped responding and has recovered. No crash.

When it fails you get 0x116. Microsoft’s reference states that an attempt to reset the display driver and recover from a timeout has failed, which happens when the driver’s threads do not finish inside the allowed window, or when TDR events pile up: the documented default there is more than five in one minute.

A blue screen is therefore the second sign of trouble, not the first. If you have had brief black screens in games for weeks, those were the recoveries that worked.

Read the file name first

The What failed line names a driver file. That word tells you who owns the problem.

File on screen	Vendor	What it means
nvlddmkm.sys	Nvidia	The Nvidia display driver. Microsoft’s worked example for this bug check names this exact file.
atikmpag.sys or amdkmdag.sys	AMD	The AMD display driver.
igdkmd64.sys	Intel	Intel integrated graphics, so a laptop or a desktop running off the CPU.
dxgkrnl.sys or dxgmms2.sys	Microsoft	The Windows graphics kernel. Read it as “whichever GPU driver was active”, not as a Windows fault.

If the screen vanished too quickly, open Event Viewer with eventvwr.msc and go to Windows Logs → System. A BugCheck event with ID 1001 lists the stop code and its parameters. Successful TDR recoveries are logged there too, under the Display source. The same technique works for every Windows stop code and what it means, and there’s more on reading a blue screen if the text disappears too fast.

Work out which situation you are in

Timing narrows this faster than any tool.

When it crashes	Most likely cause	Go to
Within days of a graphics driver update	Bad driver version for your card	Fix 2
After 20 to 40 minutes of gaming, never at idle	Heat, dust, failing fan	Fix 3
Under load on an “OC” card or with a tuning profile applied	Clocks beyond what this sample is stable at	Fix 4
Only in heavy scenes, or as frame rate spikes	Power delivery or cabling	Fix 5
Randomly, including on the desktop, after the case was opened	Card not seated, riser cable	Fix 6
Constantly, on a fresh Windows install	Driver install is broken or mismatched	Fix 1

Laptop owners can skip Fix 5 and Fix 6. Nothing inside a laptop is reseatable and the power path is fixed.

Fix 1: Clean reinstall of the display driver

A driver installed over the top of another leaves fragments behind, and those fragments genuinely cause this stop code.

Get the file from the right place. On a desktop with a separate card, that is Nvidia, AMD or Intel directly. On a laptop, use the laptop maker’s support page for your exact model, because drivers for laptop graphics are frequently customised.

Then use the installer’s clean option rather than just running it. Nvidia: choose Custom (Advanced) and tick Perform a clean installation. AMD: choose the Factory Reset option. Intel: use the clean installation option in its installer.

If the installer crashes or the driver is too broken to run it, boot into Safe Mode first. Safe Mode loads only the basic Microsoft display driver, so the vendor driver can be removed cleanly. In Device Manager, expand Display adapters, right-click the GPU, choose Uninstall device, and tick the box offering to delete the driver software.

To see what is staged on the machine, run this in an administrator Command Prompt:

pnputil /enum-drivers

Several display entries with different dates is the mess a clean install clears. Uninstall the vendor control panel suites separately from Settings → Apps and keep only the driver. They start services at boot and contribute to slow startup on otherwise quick machines.

Fix 2: Roll back after a driver update

If the crashes began within days of an update, do not reinstall the same version. Go backwards.

Open Device Manager with devmgmt.msc, expand Display adapters, double-click the GPU and open the Driver tab. Use Roll Back Driver if it is available. If it is greyed out, Windows kept no copy, so download an older release from the vendor’s archive and install it with the clean option above.

New driver branches written for new hardware are the ones to distrust. A release a few months old has had its worst bugs found by other people. Windows Update can quietly reinstall the newer driver, so check the version in Device Manager before deciding the rollback failed.

Fix 3: Temperature and dust

Microsoft lists insufficient system cooling among the hardware conditions that produce this bug check. A card that only crashes after sustained load is describing a thermal fault as clearly as it can.

Watch the temperature while you play. Task Manager’s Performance tab shows GPU temperature for dedicated cards, and vendor software adds hotspot and memory readings, which run considerably higher than the core figure.

Then look at the hardware. A heatsink packed with dust is the commonest reason a card that used to be fine no longer is. Compressed air through the fin stack, with the fans held still so they cannot spin up, fixes a good number of these. Check every case fan is turning. On a laptop, clear the vents and keep it off soft surfaces.

Throttling normally shows up as lower frame rates rather than a crash, so a card that crashes hot often has a loose cooler or dried thermal paste.

Fix 4: Undo every overclock, factory ones included

Microsoft names over-clocked components directly among the causes of 0x116. Reset the card to stock in whichever tool applied the overclock, then test again.

The part people miss: your card may be overclocked out of the box. Models sold as OC, Gaming X, Super Clocked and similar ship above reference speeds, and silicon quality varies between individual cards. A factory profile that is stable on most samples can be marginal on yours.

Testing costs nothing. Apply a core clock offset of around minus 100 MHz plus a negative memory offset in your tuning software, then play for an evening. If the crashes stop, the card is unstable at its shipped clocks. Keep the mild underclock, which costs a few percent of performance, or claim warranty if the card is young enough.

System memory belongs here too. Microsoft’s causes list includes incorrect component settings, especially memory configuration and timings. Switch XMP or EXPO off in the BIOS and run a session at stock speed. An unstable memory profile produces graphics crashes that look exactly like GPU faults.

Fix 5: Power supply and cables

Insufficient system power is another documented cause, and it has a signature: crashes only in the heaviest moments, never at idle.

Modern cards draw sharp transient spikes well above their rated average, and a power supply that coped for years can stop holding up as it ages.

Check the cabling before buying anything. Every PCIe power connector on the card wants its own cable back to the PSU, not one cable with two plugs daisy-chained on the end. Push each connector until the clip latches, and make sure a 12VHPWR or 12V-2×6 connector is fully home, since a partly seated one causes instability and heat damage. Only ever use the cables supplied with your own PSU, because modular pinouts differ between brands.

A spare power supply of adequate capacity, swapped in for a weekend, is the only conclusive test.

Fix 6: Reseat the card

Shut down, unplug the mains lead, hold the power button briefly to discharge, then remove the card. Clean dull or marked gold contacts with isopropyl alcohol and blow dust out of the slot. Refit until the retention clip clicks, and check the card sits level rather than sagging at the far end. If it is on a vertical riser, plug it straight into the motherboard for testing: tightly folded risers, or ones rated for an older PCIe generation, cause exactly this kind of intermittent fault.

The TdrDelay registry tweak, honestly

This is the advice you will find everywhere, and it deserves a straight answer.

TdrDelay sits at HKEY_LOCAL_MACHINESystemCurrentControlSetControlGraphicsDrivers as a REG_DWORD, documented as the seconds the GPU may delay the preempt request, default two. Raising it repairs nothing. It tells Windows to wait longer before deciding the card has hung, so your screen freezes for longer and the same sequence follows.

Two further points. Microsoft states that end users shouldn’t manipulate these registry keys, which exist for testing and debugging during driver development. And TdrDelay is not the key tied to this bug check: the documented one is TdrDdiDelay, which sets how long the OS allows threads to leave the driver before bug-checking with VIDEO_TDR_FAILURE, defaulting to five seconds.

There is one fair use. Long-running GPU compute, such as rendering or CUDA and OpenCL jobs, can legitimately occupy the card for more than two seconds per operation, and TDR interrupts it wrongly. That is a scheduling mismatch, not a fault:

reg add "HKLMSYSTEMCurrentControlSetControlGraphicsDrivers" /v TdrDelay /t REG_DWORD /d 8 /f

Restart for it to apply. To undo it completely:

reg delete "HKLMSYSTEMCurrentControlSetControlGraphicsDrivers" /v TdrDelay /f

If your crashes happen while gaming, skip this. You would hide the warnings and keep the fault.

What does not fix this

Suggestion	Reality
sfc /scannow	Repairs Windows’ protected system files. Vendor graphics drivers are not among them.
chkdsk	Checks the file system. No bearing on GPU behaviour.
Disabling browser hardware acceleration	Reduces GPU load, so crashes get rarer. The unstable driver or card is untouched.
Lowering in-game settings	Same masking effect, same underlying fault.
Reinstalling Windows	Reverts to Microsoft’s basic display driver, so it looks fixed until the vendor driver returns.

Microsoft’s general guidance for a repeating stop code is the sensible baseline: update or disable the driver named in the message, remove recently added hardware, and use a restore point if the fault is new.

Common questions

What does VIDEO_TDR_FAILURE mean?

It is bug check 0x116. The graphics card stopped responding, Windows tried to reset the display driver through Timeout Detection and Recovery, and the reset failed. Windows stops rather than continue with a graphics stack it cannot recover. The file named on screen identifies the vendor responsible.

Is nvlddmkm.sys a virus?

No. It is Nvidia’s legitimate display driver, and Microsoft’s own documentation for this bug check uses it as the worked example of a faulting module. Its appearance means the Nvidia driver was active when the reset failed, which points at driver version, clock speeds, temperature or power supply.

Does raising TdrDelay fix VIDEO_TDR_FAILURE?

No. It only extends how long Windows waits before declaring the GPU hung, so freezes last longer and crashes get rarer without the cause changing. Microsoft states end users should not manipulate these keys. The one fair use is long-running GPU compute work that legitimately exceeds the two-second default.

Can a failing graphics card cause VIDEO_TDR_FAILURE?

Yes, and it is the conclusion to reach last. If a clean driver reinstall, an underclock, clean cooling and known-good power all fail, and the card is stable in a different machine, the fault is likely the card. Failing memory chips on the card produce this exact symptom.

Why does VIDEO_TDR_FAILURE only happen in games?

Games are the only workload holding the GPU at full clocks, full power draw and rising temperature for a sustained period. Everything else leaves the card mostly idle. A crash pattern confined to gaming points at heat, overclock instability or power delivery rather than at Windows.

---

Read the file name on the blue screen, then clean reinstall that vendor’s driver, or roll it back if the crashes started with an update. Those two steps settle most 0x116 cases, and the registry tweak everyone recommends settles none of them.