What is the idle CPU usage of your Unix-like system setup?

[SchumannFrequency]

I'm using OpenBSD + bspwm on the Intel 12600KF.

According to top, my system is 99.9% or 100% idle when I close all apps.

It is almost always one of these two values.

What idle CPU values do you see in top when you close all apps?

What operating system and desktop/WM do you use?

 

[Easy Rhino]

99%. Usually 0.7% for the kwin compositor and 0.3% for some random misc stuff.

 

[SchumannFrequency]

ALT KWorkstation 10 with Plasma 5 is usually 99.5% idle on the DELL Latitude E6540. (Intel® Core™ i7-4610M CPU @ 3.70GHz)
This is while a weather widget and update notifications are also active.

If you use an i5-13600KF, you can normally achieve better idle values by tweaking or using another OS, although it is not particularly bad either.
I have seen Plasma 5 on some operating systems reach idle values of 94.x% and 93.x%.

On OpenBSD where I usually get 100% idle values I use bspwm + polybar + picom + sxhkd + dmenu.
music player daemon is also active when I reach those 100% idle values, but apparently it doesn't use any CPU when no music is playing.
Even though I'm using a recent AMD GPU, I still experienced tearing when I used picom (without configuring it) on OpenBSD.

I completely resolved this screen tearing (default config of picom) with the following settings in picom:

FreeBSD Desktop – Part 21 – Configuration – Compton

In this article of the FreeBSD Desktop series I will talk Compton setup – the one that does not breaks, displays everything properly and does not consume 100% of your CPU time, as unfortunate…

vermaden.wordpress.com

At the end of the article you will find a picom configuration.

If you have an AMD GPU then you can combine picom with a light window manager and with the above settings you won't have any tearing (similar experience to wayland)
The CPU will reach idle values around 100%.

The picom settings above make the app menus transparent, which looks modern and fun in my opinion.

 

[SchumannFrequency]

Calculate Linux + KDE Plasma 5 is 99.7% idle on the Intel 12600KF.

It is about 0.25% difference with OpenBSD + bspwm.

 

[johnspack]

Getting 99.8% under Manjaro, running firefox brings it down to 99.6%.

 

[mesaprotector]

98.6% idle with the only application being kitty. Almost all of the CPU use is from Xorg and picom. I'd rather have shadows and transparency than not, so I don't mind sacrificing that 1.4%.

 

[SchumannFrequency]

As an experiment, I enabled Intel hyper-threading on OpenBSD on the Intel 12600KF.
My observation is that the CPU goes to 100% idle even more often when hyper-threading is enabled.

Almost 95% of the time it is at 100% idle and 5% of the time it temporarily goes to 99.9% idle.

These are the exact values that I see almost constantly in OpenBSD.
16 CPUs: 0.0% user, 0.0% nice, 0.0% sys, 0.0% spin, 0.0% intr, 100.0% idle

I use konsole to run top.
konsole, polybar, bspwm, picom, music player daemon, sxhkd and wmctrl all run during this 100% idle measurement.

 

[SchumannFrequency]

Hardware: Intel 12700KF (stock) -- G.SKILL RIPJAWS @3600 CL18 (stock) -- Sapphire RX 7600 -- ASRock B760M-ITX/D4 WiFi -- fractal design DEFINE NANO S -- bequiet! SYSTEM POWER 10 550W -- DeepCool AG500BK ARGB -- EVO 850 500GB
Software: Artix Linux, LXQt, Mesa open-source driver, F2FS file system

Almost 80% of the time it is at 100% idle and 20% of the time it temporarily goes to 99.9% idle.
I use konsole to monitor the stats via top.
These results are very good because I see LXQt as a fully functional desktop environment.

The init system I use for Artix Linux is OpenRC.

 

[Timbaloo]

Tested btop (1 minute average) @ foot terminal @ Hyprland compositor @ Ryzen 5 Pro 5650U (edit: Oh, I run Arch btw)

No input: ~0.1% load
Constant mouse movement: ~1% load

As an experiment, I enabled Intel hyper-threading on OpenBSD on the Intel 12600KF.
My observation is that the CPU goes to 100% idle even more often when hyper-threading is enabled.

I mean activating HT "magically" doubles the number of logical cores, and since the work is staying the same (say 1% on a single physical core), the average calculated load on all cores must go down just by doubling the number or logical cores.