Stutter in KDE Plasma Desktop when using GSP firmware

[svenmauch]

NVIDIA Open GPU Kernel Modules Version

570.86.16

Please confirm this issue does not happen with the proprietary driver (of the same version). This issue tracker is only for bugs specific to the open kernel driver.

• I confirm that this does not happen with the proprietary driver package.

Operating System and Version

Arch Linux

Kernel Release

Linux 6.13.1-arch1-1

Please confirm you are running a stable release kernel (e.g. not a -rc). We do not accept bug reports for unreleased kernels.

• I am running on a stable kernel release.

Hardware: GPU

NVIDIA GeForce RTX 3070

Describe the bug

Some actions in the KDE Plasma desktop environment suffer from very noticeable stutter. For example opening the application launcher, resizing windows (especially Firefox with a loaded website) and mouse movements on the lower refresh monitors.

To Reproduce

Occurs with the package nvidia-open

Does not occur with the package nvidia and kernel parameter nvidia.NVreg_EnableGpuFirmware=0 set

Bug Incidence

Always

nvidia-bug-report.log.gz

sent to [email protected]

More Info

KDE Plasma 6.2.5-1
KWin 6.2.5-3
Wayland
1x 144 Hz monitor (main), 2x 60 Hz monitor (secondary)

I opened this issue because #693 seems like a different bug pattern and in #538 we were asked to create a new issue should it arise again.

[mtijanic]

Hey there! Yes, thank you for opening a new issue, both of those are not the right place to put this.

Unfortunately, there is not much to go on in your report above. This issue happens when there are two programs hitting the GSP at the same time and serializing on that. Under normal circumstances nothing is actively talking to GSP, so when you e.g. start a new app it has free passage to it.

However, there are tools like the KDE system monitor widget that will constantly poke GSP in the background. Again, during steady state this monitor is the only thing touching GSP so there is no contention, but as soon as a second caller comes along things can get stuttery. Do you know if you're running any such monitoring? Things like coolercontrol, ksysmon/ksysguard, nvidia-smi pmon, mangohud?

It would be helpful to provide a full list of ps aux in addition to the nvidia-bug-report.log. Feel free to send it to [email protected] instead of uploading here.

Hopefully this is not a case of some script doing while :; do nvidia-smi -q | something; sleep 1; done where the monitoring program constantly comes and goes, because then we won't be able to capture it in a snapshot. If the above doesn't reveal anything, I'll kindly ask you to run a tracing script and reproduce the problem.

Thanks for reporting!

[svenmauch]

Hi, thank you for the quick repsonse!

As far as I can tell, I don't have any monitoring tools or widgets running. I remember the stutter being much worse with version 560, but it's still definitely there and easy to distinguish from the performance with the proprietary modules. The animation when maximizing Dolphin windows is another easy way to reproduce the stutter, a program which is otherwise super smooth.

I have sent both files to [email protected] since I couldn't immediately tell how much sensitive information the logs might contain. Thank you for offering this!

[mtijanic]

I have sent both files to [email protected]

Are you sure? I'm not seeing anything related in today's emails. Can you try again? Maybe also CC [email protected] for good measure and I'll reroute if necessary.

[svenmauch]

I sent the logs again with your mail in CC. The sender address is [email protected]. Thank you.

[mtijanic]

Nope, sorry. Neither the bugs list nor personal mailbox. Not in the spam folders either. I'll check with our IT.

[mtijanic]

Okay, email issue resolved (will send details privately), got the logs. I'll look them over in a day or two and get back to you.

[mtijanic]

From the logs, I do see a thing where we are hitting the GSP a lot more than we should. We actually fixed this already while debugging #650 but it didn't make the cut for this release. It should be in one of the next 570.xx releases. I don't know if this will solve the issue fully though.

If you're willing to try, I can extract it as a patch and you can build your own nvidia_modeset.ko and see if it fixes it. Or, wait a bit for the official release and try that.

If that doesn't fix it, I would have to ask you to run a bpftrace script and reproduce the problem and send those logs as well.

Thanks again for the report and the logs!

[svenmauch]

I am travelling at the moment and could test a patch on 11.02. if that helps. Otherwise I'd wait for the release and use the old modules in the meantime.

Thank you for looking into this!

[VeilSilence]

If you're willing to try, I can extract it as a patch and you can build your own nvidia_modeset.ko and see if it fixes it. Or, wait a bit for the official release and try that.

I'm also experiencing lags and stuttering with the open NVIDIA driver (570) + GSP enabled on KWin Wayland, but I'm also facing another issue.

My "average time to render a frame" in OBS stats is much higher with the open NVIDIA driver + GSP enabled compared to GSP disabled/proprietary driver.

GSP OFF

GSP ON

Sometimes, OBS misses frames due to rendering lags in the open NVIDIA driver. The average render time constantly fluctuates between ~0.6-3.5 ms or higher, whereas with the proprietary driver, it remains stable at 0.2–0.3 ms. I'm not sure if this is the same issue as described above, so I might need to create a separate issue for it.

Can you share the patch you mentioned earlier so I can test if it fixes my issue with OBS?