Discussion: NVIDIA Display Driver (nvlddmkm.sys) DPC Latency

Status
Not open for further replies.
its an external and i forgot to turn it off thats all .......... jeeez it didnt cause any difference, just a minor inconvenience to myself...... i have stuck a post it now on the side, next to the picture of eskey on the dartboard
 
on 1809, 17763.3165 there is a package called "gpu pipeline" or similar. dont know if it exists on other major versions or not, i dont know what it does.
Originally I thought this Nvidia bug would be a W10/W11 issue and so I'd happily dive into researching something like this "gpu pipeline" thing, but I think this bug has been around for so much longer, that we probably have to look at older changes:

1) You're on W7 and have the Nvidia issue. This is surprising, but also makes sense because W7 is pretty similar to W10. What's disturbing though is that it could mean this Nvidia bug has been around for a ridiculously long time.

2) I dived into researching the history of W10, because there's all sorts of popular YouTubers and websites saying to use older versions before the timers got changed in W10, etcetera. With that in mind I looked at some older threads, like the 1809 poweruser thread on this site and they too had the same Nvidia problem back then, and they also tried to tackle it with the interrupt affinity tool. Mostly I just wanted to find someone that acknowledged this bug existing at a different point in history, which this did, so we know it's at least a few years old just from this.

3) After learning about the history of W10 I then went and downloaded a bunch of drivers from various time periods. So for example I downloaded the graphics driver from before and after the 1903 update came out, because that was an update that was really broken for gaming, and not only did Microsoft actually release a hotfix that addressed the DPC of ntsokrnl after 1903 came out, but then I would be able to try out Nvidia drivers that were created before and after this issue was discovered too. Well, all of that was fruitless, older drivers just ended up causing more problems, with errors appearing, including one of them giving me BSOD, and another crashing whenever I opened task manager. The DPC wasn't fixed in any of them either. I've always been skeptical about using anything but the latest display driver as a gamer, I think people using older drivers for performance reasons have some other issue going on that needs to be addressed instead.

4) Nvidia Watchdog Violations have been appearing for many years in Google, well before W10 was released. Most of the watchdog violations are the same bug as this thread, except it's the error that appears when your system has DPC issues so bad that it freezes the drivers and they have to restart. This ends up being yet another thing that tells us this bug is probably ancient, making it more concerning that it still hasn't been fixed.
 
Originally I thought this Nvidia bug would be a W10/W11 issue and so I'd happily dive into researching something like this "gpu pipeline" thing, but I think this bug has been around for so much longer, that we probably have to look at older changes
its a standalone package on 1809, it might be part of larger packages on 7 and 8/8.1, i dont know.
its a standalone on w7, just found it. up to you what you do but its gotta be worth a look even just to rule it out.
damn, wont upload.
 
stay_on_target_star_wars--i:14138510413981413851;d:1041398;w:240;b:000000;m:1.jpg
can you make me video show me how to do it ? :D:D:D:D:D:D:D:D:D:D
 
All updates and drivers installed
One thing we have noticed for sure, is that machines with AMD tend to have lower DPC in general. My leading theory for this is due to the custom powerplans that AMD includes with their chipset driver.

You said your power plan was on "best performance" but I'm not quite sure of what that means. Is your machine using the default balanced power plan (probably actually using the AMD balanced plan) and then in the battery slider are you dragging it to the far right for "best peformance" there in that menu, is that what you mean?
 
One thing we have noticed for sure, is that machines with AMD tend to have lower DPC in general. My leading theory for this is due to the custom powerplans that AMD includes with their chipset driver.

You said your power plan was on "best performance" but I'm not quite sure of what that means. Is your machine using the default balanced power plan (probably actually using the AMD balanced plan) and then in the battery slider are you dragging it to the far right for "best peformance" there in that menu, is that what you mean?

default windows best performance preset as it came up under windows 11 under power plan, now back to best as was before wipe etc
 
i got 2 del xp sff pc's around here somewhere and i can pick them up for 60quid each at a trusted supplier.
 
Hi
I read the topic, and did not fully understand how you consider the results of Latency Moon
Is it considered a problem if there are high readings only after rebooting for the first time for 5 minutes?
And if during games, etc., an overestimation of 1000+ occurs, is this normal behavior or under any conditions should it not be, say, higher than 500?
It's just that the first 5 minutes seem to be normal, there is nothing like that.
But while opening the application, games, I have several drivers that increase to 1000+
I'm just trying to figure out stutters in games, and trying to figure out if my problem is here.
So what is the right way to look at the delays, under any conditions should not be more than 500 or during the first 5 minutes of idle time?
 
...I read the topic, and did not fully understand how you consider the results of Latency Moon...
I'm working on writing up a guide right now that I'll try to post today if I finish, which will give people the full step by step, including how to fix general DPC issues. This Nvidia driver bug is special, because it doesn't respond to any of the usual treatments we do to fix DPC. Normally, fixing DPC is actually pretty easy once you know the how/why behind it. But this Nvidia bug is something else, which is why it has its own thread.

Here's the quick guideline for using LatencyMon though, while you wait for the guide:

1) Reboot the computer, then wait at the desktop for about 2 minutes to let it finish loading stuff.

2) Load LatencyMon and click the green start button, let it run for 5 minutes then click stop.

3) The "main" tab just shows the biggest spikes. This is usually all anyone needs to look at, because the problems are obvious when they appear. On that tab, the most important piece is the one titled "Highest reported DPC routine execution time" and will show the name of the driver that is responsible for the biggest spike during the time LatencyMon was recording.

Anything spiking over 100 can be a problem, but most people won't start noticing issues in-game or while recording audio unless it's spiking into the hundreds. The higher the spikes get, the worse all the consequences will be. In the thousands it can bring a super high-end gaming rig to its knees, even though it's getting say 300 frames per second in a game. Those frames are meaningless if the DPC is so bad that everything stutters/lags/glitches out, which is what high DPC does.
 
Все, что превышает 100, может быть проблемой, но большинство людей не начнут замечать проблемы в игре или при записи звука, если только оно не достигает сотен.
That is, even during the load on the computer, during games, nothing should exceed 100?
If during the game something spiked for 800+ - this is a problem, did I understand correctly?
It's just that it only spikes up to 1000 during load, not in idle
 
Under a heavy load the DPC will increase substantially, and that is perfectly normal, because it has much more data to process and that causes DPC to rise.

Nothing should exceed 100 during idle though. As for where things should be under load, I'm not sure, because in almost every case where DPC is a problem, once the user fixes the issues it has under idle, the issues that existed under load go away too (audio drop outs, stutters, freezing, etcetera). And so I think the reality here is that people just stop investigating further, such as trying to reduce DPC while under load.

I'll certainly look into investigating things while under load to see if I can learn anything new (once this Nvidia thing is fixed), but most people with DPC issues struggle just to fix the idle problems, and trying to get them to then follow proper protocols for testing/addressing stuff under load is only going to be more headache than it's worth, since at that point we're just making their machine marginally faster, and not solving any real problems.

There's 2 caveats to my reply here. One is that ntsokrnl will spike up even on a clean install of Windows, to about 200 or less. You can reduce this by tweaking Windows, but Microsoft has unfortunately made this "normal" to have latency spikes from this driver now, from all of the increased background activity of modern operating systems. The other obvious one is Nvidia. Due to a bug it spikes up, and also causes dxgkrnl to spike as well.
 
Under a heavy load the DPC will increase substantially, and that is perfectly normal, because it has much more data to process and that causes DPC to rise.
so, I waited a couple of minutes after reset and then ran the test for 5 minutes. ( didn't touch anything)
Result
1672785249811.png

Then, i started open browser, apps i t.d
Result
1672785329379.png
1672785352161.png
Which result is correct?
i have problem with pc or no?
 
Last edited:
Status
Not open for further replies.
Back
Top