nvlddmkm Event ID 13 and nvlddmkm.sys crashes: driver resets and TdrDelay

nvlddmkm is the service name of NVIDIA's display driver, nvlddmkm.sys, so an Event ID 13 under the source nvlddmkm is that driver writing an entry of its own to the System log, and a Display event 4101 at the same time means the driver stopped responding and Windows reset it. Raising TdrDelay fixes neither: Microsoft says end users shouldn't change the TDR registry keys, and a longer delay only keeps a hung screen frozen for longer before the reset. w0a does not write any of these values; its GPU crash recovery card deletes them, so that Windows' defaults apply again.

What nvlddmkm, Event ID 13 and Event ID 4101 are

nvlddmkm.sys is the kernel-mode part of NVIDIA's display driver: the file's own properties describe it as "NVIDIA Windows Kernel Mode Driver", and Windows runs it as the service nvlddmkm. The events the driver writes to the System log carry that service name as their source.

An event ID means something only together with its source. Under the source Display, 4101 is Windows' own record of a driver reset. Other sources reuse the number: on the PC these lines were checked on, two dozen other sources of the System log, Tcpip and Dnscache among them, define a 4101 of their own. Under nvlddmkm, 13 is a number NVIDIA's driver chose for itself.

Windows cannot put a sentence to that 13. On the PC this guide was checked on, with NVIDIA driver 621.01, the nvlddmkm source has no message file registered, so Event Viewer says it cannot find the description for the event ID and then prints whatever text the driver attached. Microsoft's TDR documentation does not list NVIDIA's IDs either, so this guide does not guess at a meaning: the driver's text and the time are what you can go on.

Display 4101 is spelled out. Its text reads "Display driver … stopped responding and has successfully recovered.", with the driver's name in the gap.

What a driver reset is

Microsoft calls the mechanism timeout detection and recovery (TDR). When the GPU can neither finish nor preempt its current task within the timeout, "the OS diagnoses that the GPU is frozen", and Windows resets the graphics card "to prevent the entire system from becoming unresponsive". The default timeout is two seconds. After a successful recovery, Windows shows "Display driver stopped responding and has recovered." and logs that message in Event Viewer.

A reset does not always end there:

• Repeated resets end in a blue screen. By default, Windows bug-checks the PC "on the sixth (or more) GPU hang when it detects that five (5) or more GPU hangs (0x117) and subsequent recoveries occur within one (1) minute."
• A reset that fails is a blue screen at once. Bug check 0x116 VIDEO_TDR_FAILURE "indicates that an attempt to reset the display driver and recover from a timeout has failed". Microsoft's own sample analysis of that stop code names nvlddmkm.sys as the faulting module.
• The game can stay black. Microsoft notes that "Some legacy DirectX applications might just render black at the end of this recovery, which requires the end user to restart these applications."

Find the events in Event Viewer

1. Press Windows + R, type eventvwr.msc and press Enter.
2. In the left pane, open Windows Logs and select System.
3. Click the Source column header to sort the list; the Display and nvlddmkm entries now sit in blocks of their own.
4. Select an entry and read the General tab below the list, together with its time and its Event ID.

PowerShell gives the same list in one line, newest first:

Get-WinEvent -FilterHashtable @{ LogName = 'System'; ProviderName = 'Display', 'nvlddmkm' } -MaxEvents 30 | Format-Table TimeCreated, ProviderName, Id -AutoSize

If PowerShell answers that no events were found, neither source has an entry left in the System log, and this page is not your trail: Why did my game crash? covers the other records.

How to read what you found

Read the times side by side. A Display 4101 at the moment the picture went black and came back is a reset that worked. An nvlddmkm entry in the same minute belongs to the same incident. Several 4101s within one minute followed by a restart match the five-in-a-minute rule above, and the dump of that blue screen is where Why did my game crash? picks up.

What to check, one change at a time

Microsoft's resolution for 0x116 starts with the driver, then lists hardware: "Over-clocked components", "Incorrect component compatibility and settings (especially memory configuration and timings)", "Insufficient system cooling", "Insufficient system power" and "Defective parts". Test after each step, so you know which change mattered.

1. Reinstall the driver cleanly. Microsoft: "You may need to install the latest updates for your display driver, so it properly supports the TDR process." In the NVIDIA app, open Drivers, choose Reinstall for the installed driver, pick Custom installation and tick Clean installation, which the app says "Restores all driver settings and profiles to default". Clean Install for Stable 1% Lows walks through a full driver reinstall, Display Driver Uninstaller included.
2. Take overclocks off the graphics card. Turn on Debug Mode under System > Advanced in the NVIDIA app; the app says it "Removes all currently applied overclocking (user provided and/or factory provided) from the GPU until system restart". If you applied an offset with w0a, Reset to Default on its Graphics Card page puts the stock values back.
3. Test the memory at its default settings. Memory configuration and timings are on Microsoft's list, so an XMP or EXPO profile or hand-tuned timings are worth one test without them. Is your RAM at its rated speed? shows how to check the profile and test the modules.
4. Check cooling and power. Watch the GPU temperature under load with CPU and GPU temperatures while gaming; if the PC also switches off by itself, PC restarts or shuts down while gaming covers the power side.
5. Update the game. Microsoft's checklist asks for "any graphics intensive applications (such as games)" to be "fully patched".

Why w0a never writes TdrDelay or TdrLevel

Changing the TDR delay means raising TdrDelay above its default; switching TDR off means setting TdrLevel to 0. Microsoft's TDR registry key reference is plain about both: the keys are for "testing or debugging purposes only during the driver development process", and "End users shouldn't manipulate these registry keys."

What the two values do follows from the same page. TdrDelay "is effectively the timeout threshold", two seconds by default: a hung card is reset only when that time runs out, so a bigger number does not stop the hang, it keeps the screen frozen for longer before the reset. TdrLevel 0 is "Detection disabled": Windows no longer notices the hang, so nothing resets the card.

w0a writes no TDR value: TdrDelay and TdrLevel are on its banned list, which even a restore from a backup cannot get past. The reason is a real machine: older w0a builds raised TdrDelay to 10 seconds, and there these keys froze CS2. What w0a does is the opposite.

The GPU crash recovery card on the Fix page deletes TdrLevel, TdrDelay, TdrDdiDelay, TdrLimitCount and TdrLimitTime under HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers. With the values gone, Windows runs on its defaults: recovery on, a 2-second timeout, 5 seconds for TdrDdiDelay, and five resets allowed within 60 seconds. Type tdr into the page's search box to find the card. It asks for a restart when it deleted something, and says so when there was nothing to delete.

The card does not stop a GPU from hanging. It puts Windows' own rescue back, and the cause is still one of the checks above.

The Stutter Detective counts the Display 4101 events of the last 24 hours, and warns when TdrLevel 0 has switched recovery off.

Frequently asked questions

What does nvlddmkm Event ID 13 mean?

It is an entry NVIDIA's display driver wrote under its own source, and Windows has no description of its own for it. Read the text the driver attached on the General tab, and look for a Display 4101 in the same minute: if there is one, the driver was reset, and the checks above apply.

Is nvlddmkm.sys broken if a blue screen names it?

Not necessarily. Microsoft's own example of VIDEO_TDR_FAILURE names nvlddmkm.sys, and the causes it lists run from the driver to overclocking, memory settings, cooling, power and defective parts. For a driver named in a bug check, Microsoft's advice is to check with its manufacturer for driver updates.

Should I raise TdrDelay to stop the crashes?

No. Microsoft says end users shouldn't change it, and a longer TdrDelay only makes Windows wait longer before it resets a hung card. If an older tool raised it, the GPU crash recovery card deletes it.

Can I turn TDR off with TdrLevel 0?

You can, and that is the problem: Microsoft lists 0 as "Detection disabled". Windows then does not detect a hang at all, so it does not reset the card, and the reset exists "to prevent the entire system from becoming unresponsive".

Why does the game stay black after the screen comes back?

Microsoft notes that "Some legacy DirectX applications might just render black at the end of this recovery, which requires the end user to restart these applications." Restart the game; if the resets keep coming, work through the checks above.

Related

• Why did my game crash? Reliability Monitor and Event Viewer
• Clean Install for Stable 1% Lows
• CPU and GPU temperatures while gaming: what is normal
• Is your RAM at its rated speed? XMP, EXPO and dual channel
• PC restarts or shuts down while gaming: Kernel-Power 41