Until now, I've automated a command which disables desktop effects when playing fullscreen games, in order to improve performance. However, I'm starting to wonder if I should also kill plasma-desktop to improve performance even further. I use an own script to launch full screen games, and could easily add the command there to handle it.

I'm curious how much Plasma affects performance of games, even when they're ran in full screen. Has anyone ever measured, and does any other gamer kill Plasma when playing high-end games? From my knowledge, the computer's resources are still used to render the desktop, especially since I have many plasmoids which update constantly... so I'm tempted to go with this idea.

For the time being, I already adapted my script to kill Plasma before launching games and start it back up after the game closes. The performance improvement is visible in at least two games! FPS isn't improved much, but everything loads and works slightly faster. I was aware some plasmoids consume CPU, but didn't know it would visibly affect games. Certainly I'll keep to this practice, but would still like to know what others think and how Plasma affects their game performance.

One probable reason for the "increased performance" could be less context switches (which are "expensive" operations to perform), or alternately, the lack of disk activity which Plasma Desktop triggers - through writing to log files such as ~/.xsession-errors for instance.

bcooksley wrote:One probable reason for the "increased performance" could be less context switches (which are "expensive" operations to perform), or alternately, the lack of disk activity which Plasma Desktop triggers - through writing to log files such as ~/.xsession-errors for instance.

Makes sense. The performance it seemed to reduce was small hiccups here and there, as well as content loading slower before a map start. I have a photo frame slideshow widget, many system monitor widgets (updating every 1 second), RSS widget and more. Those are likely the cause for lower performance.

I wonder if Plasma can be configured to disable widgets while certain processes are running. The same way KWin can block compositing while listed processes exist. Would be an useful feature, and I wouldn't have to fully disable Plasma in that case.

Something else that might be worth trying is to use the SIGSTOP and SIGCONT signals to essentially freeze and unfreeze the plasma-desktop process, thereby gaining the performance benefits (reducing background CPU load) without having to re-launch it and wait for it to come up after you come back from the game (other than perhaps loading the process back from swap if it got swapped out of memory while you were playing the game).

Please note that "freezing" the Plasma Desktop process may result in unexpected freezes in other components - particularly in those related to notifications.

I must say I somewhat doubt this claim. Did you actually do statistics on how much it improves? i.e. did you do this at least five times: start the game without plasma running, stop the load time with a stopwatch, then reboot your computer; start the game with plasma running, stop the load time with a stopwatch, then reboot your computer; compare the two averages and deviations with each other and verify there's indeed a significant improvement?
In PowerTop, you can see that plasma-desktop in idle state has about 7 wakeups per second. I'd assume that number is related to the number of context switches required for this application... and that sounds barely relevant. The exception would be if you have very little system memory, then the extra maybe 70MB from quitting plasma-desktop could be relevant.

From my experience load times are highly subjective and I myself have more than once thought to have an improvement in a load time when measuring it indeed showed there was none.

Same goes for desktop effects btw, KWin should (I think) automatically turn this off for fullscreen windows (to be honest I can't explain what it does exactly, but it does something to make compositing not affect fullscreen windows, basically). From my tests, it also really doesn't matter whether compositing is turned on or off -- in recent versions of KWin, that is. Three years ago, or with compiz, I do remember a huge impact on performance.

Cheers,
Sven