Maxing out single core with v9+.

[RobertZenz]

I'm seeing something similar to other issues already reported but unsure whether to open a new issue or not. I'm running a version from the end of 2020 with an NVIDIA 1050Ti, official drivers, experimental backends/glx and it works as expected. However, with newer versions (9+) I'm seeing a large CPU usage whenever I'm watching a stream with mpv (maxing out one core, I believe). Configuration is unchanged.

I've disabled vsync (because I've enabled full composition in the NVIDIA drivers) and also disabled unredirect for that window so that compositor does not get enabled "just" because of the stream.

I guess I could do a bisect to try whether I can find the commit with which it started, is there something else I should keep in mind or could do?

[RobertZenz]

I've retested with v11.2 and the high CPU usage is still there, however, it seems related to disabling V-sync.

I'm currently running picom with disabled V-sync because I've enabled forced composition in the nVidia drivers ("Force composition pipeline") with my, now, 3060. If I run v11.2 with this configuration it will use up/max out a single core, if I hand it --vsync it will behave like the old versions and the CPU usage will be neglectable. The old version is running with the (then) experimental GLX backend, the new version with EGL.

That leads me to believe that maybe this is a bug in the old version I'm running, as it seems like the old version performs some frame limiting even though V-sync is disabled. The fact that it uses up one core seems logical, as without V-sync it should output frames as fast as possible, but what confuses me is that the old version does not do that.

Anyone got some new ideas regarding that?

[RobertZenz]

I've just tried v12-rc1 (NVIDIA RTX 3060, 555.58.02, "Force Composition Pipeline" enabled) with the egl backend and, with the exception of the fully transparent windows, it does work and does not show the high CPU usage. It seems to be the same CPU usage as v9 had. So this seems to be fixed.

However, memory usage seems to have roughly doubled, from ~41MB to ~98MB, compared to v9 if that is of any interest.

[RobertZenz]

However, memory usage seems to have roughly doubled, from ~41MB to ~98MB, compared to v9 if that is of any interest.

After a reboot, this is gone and v12-rc1 is now sitting at ~41MB again. Curios, maybe my system was confused by previous picom starts/kills or something.

So I'd say all of this solved, thanks to everyone involved in development!