Created attachment 174146 [details] Quick plot of the two frametime traces SUMMARY When using an external screen with my laptop, I see random lag spikes on the image of this external screen, roughly once per minute. During these lag spikes, the frame time goes from ~16 ms to 150-300 ms for a few frames, before going back to normal. While testing, I saw that this specific problem does not appear with a Windows install, nor with latest Mutter (47.0). So it seems it is coming from something Plasma specific, and not from a problem with my hardware or with Nvidia linux drivers in general. SOFTWARE/OS VERSIONS KDE Plasma Version: 6.1.5 KDE Frameworks Version: 6.6.0 Qt Version: 6.7.2 Kernel Version: 6.10.10.arch1-1 (64-bit) Processors: 16 × 11th Gen Intel® Core™ i7-11850H @ 2.50GHz Memory: 14.8 Gio of RAM Graphics Processor: Mesa Intel® UHD Graphics / NVIDIA RTX A3000 Laptop GPU/PCIe/SSE2 Nvidia driver version : nvidia-open 560.35.03 ADDITIONAL INFORMATION I used mangohud with vkcube (using Nvidia GPU) running maximized on the external screen to capture the frametimes traces attached I have tried multiple configurations in order to better determine the source of these lags: - X11/Wayland : I see the same type of lag spikes. Wayland has more problems still with Nvidia, so I continued to investigate only with X11 - Hybrid graphics/Nvidia only (BIOS configuration) : this specific problem disappears in Nvidia only configuration (but the frametime is more chaotic ...) - Output resolution 1080p/4K : no change - Laptop output DP/HDMI : no change - I tried multiple screen configurations with different external screens, with no impact on the lag spikes - KWin/Mutter : Lag spikes disappear in Mutter I don't have any ideas on where/how to investigate further, so I'm reporting the issue here
Created attachment 174148 [details] Frametime data in KWin
Created attachment 174149 [details] Frametime data in Mutter
KWin doesn't influence application's synchronization with the screen's refresh rate on Xorg, so even if this doesn't happen in other environments, it's still a driver bug, which you can report at https://forums.developer.nvidia.com/c/gpu-graphics/linux Note that if you have a widget or something that monitors GPU usage or something like that, those generally query once every second, and there's a known driver bug that makes that querying cause severe performance issues when the GSP firmware is used.
If someone else stumbles upon this report, I have found the origin of the problem : I have installed a plasmoid (https://github.com/davidhi7/ddcci-plasmoid) to manage the brightness of my external display through DDC/CI. This plasmoid polls every minute (the frequency of the spikes I saw, my graphs should read "time (min)" instead of "time (s)") for the presence of DDC/CI compatible screens, and this polling seems to heavily affect framerate .