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

Status
Not open for further replies.
It would be interesting a step by step on how to test this in LatencyMon. For example, do you just leave the tool running without any other program open? Do you disable the antivirus or not? Modify the power plan? What power plan do you use to run the tests? I use the Balanced plan even in games and I used it in my last test with LatencyMon without having any problems, however I noticed that the latency of some here is much lower.

I think it's important to have a standardized test methodology that anyone can reproduce.

If I have time I will try to do a clean install of the unmodified official Windows 10 22H2 ISO and the custom ISO with NTLite to compare the results. But I suspect I don't have any problems with latencies here.
 
Well, hello, club. I've never wanted to belong less to a club than this thread *sigh.
1672067668376.png
I think this pretty much speakes for itself although I've only been running LatencyMon for a minute. All other drivers are below 0,3 ms. This is my new PC I've built about a month ago. I'm quite frustrated because I'm a recording guitar player and I have to set my latency so high on my audio interface because of pops and clicks that I'm literally unable to record anything.
I hope we'll eventually have a solution because this is ridiculous and at this point I'm close to ripping this 3080 out of my PC and just going AMD.
 
Last edited:
Power mode in NVIDIA CP is in High performance.
oskuro said:
I just had again that electric sound interference and PC totally froze again. After 3-4 tries restarting it, entering BIOS... I could boot Windows again.
So what I thought was an issue caused by RAM maybe it´s because this damned i9 12900K messing around with Asus mobo and my GPU?.

This is a nightmare. Never ever had these problems with any of my previous PCs.

1. New Test: Removed GPU and RAM in slots 1&3 (were in 2&4 before). Moved also M.2 from Slot 1 (above 1st PCI for GPU) to 2nd M.2 slot (Z690-F Asus Strix). Had again the freezing while listening Spotify. Screen ended this way:

https://i.ibb.co/5FjJ8Q6/Sin-t-tulo.png

2. Now I will test with just 1 RAM in slot "1", no GPU and M.2 same place as before (2nd M.2 slot). I will turn on Spotify and see what happens. With LatencyMon in background as well.

These are the conflicts I have without the GPU and just 1 RAM:

View attachment 8813

VS the ones I had before:

index.php


10 mins running LatencyMon with Spotify, using Chrome and normal things... and no freezings:

View attachment 8814

Mmmmm I will test again with 2 modules of RAM (1&3), but If I get the freezing again as before... does it mean that maybe RAM´s defective?. I still didn´t put again the GPU, as I want to test the freezings first without it.
So it´s not a GPU matter (DPC spikes are of course when I connect it but that again it´s related with Z690 I guess... damned Asus...).
Click to expand...

Bah, I thought it was already solved but again with the damned freezings. I think this is the last time I buy a damned Asus Mobo.

Theses issues just happen with i9 12900K (alder lake series) + Asus Motherboards?. Or also another brands?.

It´s a joke you buy a new PC and you have to be touching 9999999999999999 things to not have freezings, ridiculous.
 
oskuro said:
Power mode in NVIDIA CP is in High performance.


Bah, I thought it was already solved but again with the damned freezings. I think this is the last time I buy a damned Asus Mobo.

Theses issues just happen with i9 12900K (alder lake series) + Asus Motherboards?. Or also another brands?.

It´s a joke you buy a new PC and you have to be touching 9999999999999999 things to not have freezings, ridiculous.
Click to expand...
I'm on a Maximus Z790 Hero and 13900K. Asus confirmed.
 
Whilst I see a lot of griping about everything else under the sun. maybe we should take a step back or 2......

1) I have an asus mobo, and I DO NOT experience intermittent sound issues , lags or delays
2) I have an Nvidia card ....... and i do not experience issues.
3) richaxes i record audio too (bass, guitar, keyboards and vocals ) on mine, but i don't have the issues you have. ( but i'm on amd not intel )

under Nvidia CP I have power set to adaptive, maybe that could be why I don't experience it, I honestly don't know. The solution could be anything from Nvidia sorting it, to MS sorting it, but no-one is willing to fully fess up and release a patch.

AeonX I have run tests with and without AV running, and posted them up here, windows power plan set to high performance on both so i doubt its power plan related

Overall it is an issue some of us are having,not all of us, but some of us. I don't want to discourage any help other people may be able to provide, or to rubbish anyones input, but what really isn't that helpful is just coming in and saying ' i got this problem' and not really providing the full picture. As Hellbovine has suggested and i do think it isnt being unreasonable, is a 5 min latmon run. It gives time for all process or whatever to be seen instead of posting a 1 minute one where there can always be inconsistencies.
That way, with posting a 5 min latmon results thingy.. maybe we can from ALL systems involved be able to see if there is a direct link or causality, then everyone playing guessing games. If it were that easy to sort out, this thread would be closed with just the solution. Different users use their computers for different things, so sadly it will take time to see if any solution can be found. Patience is greatly appreciated as we all work through this together and help each other out. that's what these forums are here for
 
Taosd said:
Whilst I see a lot of griping about everything else under the sun. maybe we should take a step back or 2......

1) I have an asus mobo, and I DO NOT experience intermittent sound issues , lags or delays
2) I have an Nvidia card ....... and i do not experience issues.
3) richaxes i record audio too (bass, guitar, keyboards and vocals ) on mine, but i don't have the issues you have. ( but i'm on amd not intel )

under Nvidia CP I have power set to adaptive, maybe that could be why I don't experience it, I honestly don't know. The solution could be anything from Nvidia sorting it, to MS sorting it, but no-one is willing to fully fess up and release a patch.

AeonX I have run tests with and without AV running, and posted them up here, windows power plan set to high performance on both so i doubt its power plan related

Overall it is an issue some of us are having,not all of us, but some of us. I don't want to discourage any help other people may be able to provide, or to rubbish anyones input, but what really isn't that helpful is just coming in and saying ' i got this problem' and not really providing the full picture. As Hellbovine has suggested and i do think it isnt being unreasonable, is a 5 min latmon run. It gives time for all process or whatever to be seen instead of posting a 1 minute one where there can always be inconsistencies.
That way, with posting a 5 min latmon results thingy.. maybe we can from ALL systems involved be able to see if there is a direct link or causality, then everyone playing guessing games. If it were that easy to sort out, this thread would be closed with just the solution. Different users use their computers for different things, so sadly it will take time to see if any solution can be found. Patience is greatly appreciated as we all work through this together and help each other out. that's what these forums are here for
Click to expand...
Okay, so please let me how can I exactly be helpful and I'll happily oblige.
Edit: OK, you did, I'll post it soon.
 
Taosd said:
under Nvidia CP I have power set to adaptive, maybe that could be why I don't experience it, I honestly don't know.
Click to expand...
Interesting. Adaptive is not full performance (and priority). Therefore the driver doesn't try hard to beat out other apps.
What happens when you crank it to max perf?
 
garlin said:
Interesting. Adaptive is not full performance (and priority). Therefore the driver doesn't try hard to beat out other apps.
What happens when you crank it to max
Click to expand...
changed Nvidia CP power under both tabs from adaptive to prefer maximum performance (power management mode under global - Prefer Maximum Performance. and Program settings tab - prefer maximum performance)

under youtube playback, 0 difference . no pops, crackles or anything abnormal
Studio software - 0 difference. no pops, crackles or anything abnormal
in game - 0 difference. no pops, crackles or anything abnormal

to be honest no difference, but wasnt having the issue originally,

but everything that can be ticked off, or may be successful for someone else is worth that shot
 
Seems to be a more common issue on Intel, have you tried disabling Spectre Meltdown mitigations? I don't know if this affects recent processors but the old ones suffer a lot from performance loss with this enabled. I just remembered that I always turn it off.

Code: 
reg add "HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management" /v "FeatureSettingsOverride" /t REG_DWORD /d "3" /f
reg add "HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management" /v "FeatureSettingsOverrideMask" /t REG_DWORD /d "3" /f
or
Code: 
Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management]
"FeatureSettingsOverride"=dword:00000003
"FeatureSettingsOverrideMask"=dword:00000003

A reboot is required after applying the reg values.
 
I have the same issue with the Nvidia driver.

My LatencyMon results: https://pastebin.com/EfNxmbXb

PC Type: Desktop
Operating System/Version: Windows 11 Pro 22H2
Processor: Intel i9-13900K
Graphics: (nVidia) Asus GeForce RTX 3080 STRIX OC - GPU is not sharing IRQ as far as I can tell
Driver Type: Game-Ready
Driver Version: 527.56
DPC Spikes: as included in the results

I'm using multiple monitors, two M.2 SSDs (Samsung 980 Pro 2Tb and Kingston A2000 1Tb) and no SATA devices at all.
 
Last edited:
PC Type: Desktop
Operating System/Version: Windows 10 LTSC 64-bit 21H2
Processor: i7 9700k (Coffee Lake)
Motherboard: z370 hd3
Graphics: Gtx 1070 - GPU is sharing IRQ 16 with High Definition Audio Controller
Driver Type: Game Ready
Driver Version: 526.86
*The preset/reg tweaks removes and disable a lot of stuff, it's better to use them on fresh install for testing.
edit: adding new tests with clean install with no modifications:
Image1 - with nvidia driver as is.
Image 2 - manual cleanup of the driver, only physx and the base driver installed.
*clean install for each of the image1/2 tests.
 

Attachments

  • Image.jpg
    Image.jpg
    182.1 KB
  • Preset.xml
    88.8 KB
  • Regtweaks.reg
    62.6 KB
  • Image1.jpg
    Image1.jpg
    253.7 KB
  • Image2.jpg
    Image2.jpg
    296.6 KB
Last edited:
*IMPORTANT UPDATE*
I want to thank everyone for participating here, I never expected this thread to grow so big. I genuinely appreciate all the activity this has received and I hope we will find a solution. I have to take a moment though to acknowledge a problem we're starting to run into. This thread is really skyrocketing in activity lately and it's becoming information overload as a result. In order to prevent it from being locked before a solution is found, we all need to focus on streamlining some things as we move forward in the thread. Please read through the guidelines listed below, and try to apply these principles in the future, if you aren't already doing so. Thank you guys!
*IMPORTANT UPDATE*

AeonX talked about testing methodology, and I couldn't agree more. I didn't expect finding a fix to take 6 months though, otherwise I would have laid much better groundwork for this thread earlier so that we were all on the same page in testing. I think the following rules will help to keep this thread on track as we work towards a solution, and should also make testing more consistant:

RULE #1: RELEVANCE
The goal of this thread is to fix the Nvidia DPC bug. This thread is not intended to solve everyone's general DPC issues. The reason the Nvidia bug is special is because it's affecting a huge population, and it doesn't respond well to any traditional fixes so far.

If you need to solve general DPC problems from sources other than Nvidia, then those can almost always be easily resolved by following the steps I laid out in my general DPC quick fix guide (link1) from earlier. For further assistance, you need to either wait for me to post a general DPC guide so you can use that new thread to discuss those issues in, or look for assistance in other sites/forums that are more focused on audio production or gaming. It's not that I don't want to help people here, but this thread is simply too busy already with the Nvidia bug, and it's just not efficient to try and solve everyone's other issues while we are still struggling to find a fix for this special bug.

RULE #2: LIKE BUTTON
Due to the number of members involved in this thread now, we have to eliminate most cross-talk. What does this mean? Moving forward we have to cut down on stuff that could be construed as goofing around, and replies that don't directly add to the discussion. Examples of this would be things like, "Man I have this issue too, I don't know what to do!" Instead, we all need to make use of the "Like" button to convey these sentiments, since they do not take up extra thread pages like a text reply does.

Basically, we need to let the "Like" button do a lot of talking for us if possible. By all means though, if you feel that a question you have hasn't been asked or answered yet, then ask it. If someone makes a post and you don't agree with some of the information, it's okay to reply and correct it. I'm not trying to silence people at all, only to eliminate posts that aren't really discussing relevant things or contributing anything new. Please don't use this rule to cause uneccessary arguements, we don't need to point out right away if someone made a low quality post, unless it continues to be a problem from them. We're all still human and make mistakes!

RULE #3: NEW POSTERS
It's okay to post here acknowledging that you have the same issues, and even now this still helps us because then we can see your hardware/software information and we can look for similarities/differences which may help us find a solution. Whether you do or don't have DPC issues, it still helps us if you post either way. There is a template (link2) that you can fill out.

Before posting please take the time to read through all 20+ pages in the thread so far. I realize it's a lot to digest and will take some time, but quite a few of us have already done this, and have also put in countless hours into researching and troubleshooting this bug, and it's not fair to the most active troubleshooters to have them rehash the same information over and over again.

RULE #4: TESTING
Testing methodology is really important, and it's unfortunately something that everyone does differently. There are right and wrong ways to test things, and some of the required testing aspects for this thread are listed below:

- Use the latest version (7.20) of LatencyMon and be sure to run it as administrator.

- LatencyMon screenshots need to be for 5 minutes. It's okay to be a few seconds over, but that's the only exception.

- LatencyMon should be ran only after a reboot, not while games are running or with a browser open, etcetera.

- Don't post a proposed solution without also including a before/after LatencyMon screenshot of the fix.

- Don't post multiple solutions at once, spend the time narrowing it down to the 1 tweak that actually fixed it.

- If you have an idea but cannot test it yourself, please suggest it, just try not to mention anything already discussed.

Example of the list above in action:
If you are posting a proposed solution you are expected to have spent time testing it. No matter what kind of tweaking you're doing on a computer, the only proper way to know if something is a "fix/improvement" is if it can be replicated through precise steps, and it works every time. If you change 20 settings all at once and suddenly things are "fixed", please do not post the list of 20 things. Instead, spend the time going back and verifying if you can replicate your success. In other words, undo all the tweaks, then do one tweak at a time, running LatencyMon after each, to see which item was truly responsible for your success. If you really want to put it to the test, do a clean install of a default Windows and see if the fix also works there.

This rule may sound ungrateful at first, but the harsh reality is that most of the time people only *think* they found a solution, but because of a lack of testing, or poor testing methods, they really didn't find a solution. Then other people have to spend a ton of time methodically going through a list of 20 things, only to find that none of them work, or even worse this information spreads around the internet and joins the other piles of bad tweaks and misinformation that gamers blindly apply to their computers. I'm currently spending more time managing this thread because of issues like these, than I'm able to spend time exploring solutions of my own.

As a final point of clarification, some people may come up with good ideas that they cannot test themselves. In those cases, please of course give us ideas to explore, just try your best to not post anything that has already been discussed. Remember, we're not trying to be rude with all these rules, rather it's about efficiency--finding a solution as fast as possible, as easily as possible. Otherwise, this thread will get closed if it fills up with a bunch of bad solutions and people posting unrelated DPC issues.

Note: also updated the original post on page 1 to add various checkpoints in this thread, including this one.
 
Last edited:
ive sat here and watched LaMon with windows idling and all seemed fine them blammo, a spike. start LaMon and do stuff for 10 minutes(whats 10 minutes after all?) as it is far more, oh damn, whats the word? its what you would normally do.
 
Clanger said:
ive sat here and watched LaMon with windows idling and all seemed fine them blammo, a spike. start LaMon and do stuff for 10 minutes(whats 10 minutes after all?) as it is far more, oh damn, whats the word? its what you would normally do.
Click to expand...
Your case is interesting because you're on W7 while most of us are on W10/W11...Are you implying that your Nvidia can sometimes go for 5 minutes without spiking? If so, we should investigate that. I've never had a 5 minute run where it didn't spike though.

I think this thread is starting to blur the line between general DPC issues and Nvidia DPC, and so it's getting really muddy. It certainly won't hurt anything if someone wants to record for *longer* than 5 minutes, we just can't have people posting 1 and 2 minute screenshots because those are useless for the reasons you said.

The other thing to consider too, is if we all post the same duration, it makes the data easier to compare. What I mean is, let's say I'm posting 5 minute screens and you post 10 minute screens, well when I look at the "DPC Count" column, now I have to plug in all our numbers into a math formula to try and divide the counts between the time duration, rather than being able to straight compare them, which slows everything down.
 
i havnt actually timed myself but 1 minute is nothing, i'd certainly leave LM running for a while. a lot of people are still using 7.
 
Clanger said:
ive sat here and watched LaMon with windows idling and all seemed fine them blammo, a spike. start LaMon and do stuff for 10 minutes(whats 10 minutes after all?) as it is far more, oh damn, whats the word? its what you would normally do.
Click to expand...
Most likely something started up in the background that caused the random issue. If it happens later on in the run 3.5-4.5 just stop and restart. If it was something in the background then it wouldn't do it again
 
LMon usually reports im ok for real time audio stuff but if i do some of those platform tick type bcdedit tweaks then i run into trouble, LM complains and the audio can be very glitchy. audio is more important than frame rates.
 
Catching up on recent posts. People brought up some good points that I'd like to expand on:

LATENCYMON DATA
LatencyMon will give a text summary report next to some sections, saying things like "Your system is fine". These sentences are probably responsible for a lot of confusion, and now I'm kicking myself for not recognizing this problem earlier. The text summary in LatencyMon should be ignored entirely. It is arbitrary and just a guideline for the uninitiated, any spikes less than 2000 will be "fine", while spikes between 2000-4000 will report "doubtful", and spikes above 4000 will be "problematic" in the text summary. The devs should really lower these numbers in a future version.

In the audio production and gaming communities, excellent DPC is defined as spikes under 100. On Windows XP for example you could easily have an average DPC of less than 10 microseconds and wouldn't see a spike over 50. Anything spiking over 100 can be problematic, and once you reach about 400 or more you can start to easily see/hear the consequences of those spikes. The higher the spikes go, the more frequently they happen, and the more sensitive the user is, the more noticeable these consequences get.

A wrench thrown into all of this is that modern Windows now have extremely high default DPC. Windows 10/11 for example have Ntsokrnl spiking up to 200 and that's "normal" because DPC has not been a focus of Microsoft's. Because of issues like this, the average DPC in modern Windows is much higher too.

NVIDIA BUG
Something that keeps coming up in this thread is a sentiment that not everyone has this Nvidia issue, but I think the opposite is true and almost everyone does have this bug. The average spikes are between 500-800, and in computers that have "general DPC problems" this will compound the delays and make it spike into the thousands. How bad the problem is really depends on how bloated a user's OS is (if they tried to reduce background overhead by tweaking things or not).

At the time of this post, four people in this thread have said that they don't have the Nvidia issue (defined as Nvidia driver not spiking into the hundreds). However, there's a problem with each success story that makes the data unreliable, with the main issue being that none of the users have been able to replicate the steps they took to solve it, or nobody else can replicate it. I'm not saying these people didn't find a fix, but until it can be replicated then we don't have a solution.
 
Last edited:
qou said:
...*The preset/reg tweaks removes and disable a lot of stuff, it's better to use them on fresh install for testing.
Click to expand...
So just to clarify, you have no Nvidia spiking, and the highest spike you got on a system that has all drivers installed, including Nvidia, was less than 200? If you answer yes to that, then the next question I have is, does your same machine still get similar results on a clean install of a *default* Windows (no modifications) with the Nvidia driver?
 
Last edited:
Status
Not open for further replies.
Back
Top