Over the years I've seen a lot of poorly made comparisons of Linux Desktop Environments in terms of system usage and in this article I want to make things right once and for all.

Let's start with a methodology which absolutely needs to be defined because we want reproducible results, not something which is random and user dependent.

For testing Fedora spins are used - you can find download links below in the comparison table. Fedora normally offers quite fresh if not the freshest software - we need something which is relevant as of now, December 2022, not December 2012.

Post installation the system will be fully upgraded and rebooted - that's how users will run it anyways.

Tests will be performed under a VM, VirtualBox 6.1.40. Why? People's hardware varies greatly, it's impossible to predict whether your system has a fast dedicated GPU or a built-in one. This could result in different RAM/VRAM usage patterns. VirtualBox 7.0 will not be used as it's still considered beta quality software.

A VM will get a 16GB disk drive, a four-core vCPU, 128MB of VRAM and 4096MB of RAM running at 1920x1080@24bit resolution. SWAP will not be used as it behaves unpredictably.

Gnome and KDE will be tested under Wayland and Xorg. All other DEs will be running under the Xorg server, i.e. in pure X11 mode.

To assess the system RAM use without a desktop environment running, the system will be booted into IceWM which has a very miminal footprint yet offers a sort of usable graphical environment.

Once logged in, default file manager and terminal emulatar applications will be started and that's it. Most users use both a lot, while it's hard to assess what other applications are common.

Disk buffers will be flushed using echo 3 > /proc/sys/vm/drop_caches. Why? Because they may never be re-used. We need to assess the actual RAM use.

What none of the existing/previously made comparisons have taken care of is disk IO use for DEs. I will use VirtualBox Session Information feature to get this data 60 seconds after booting into a particular desktop environment.

Desktop environments are used as is without any modifications or additions. For instance, KDE Plasma Add-ons (plasmoids) inflate RAM usage even further. KDE is already the biggest resource hog.

Without further ado and half an hour youtube videos full of nothing here's the data in MB (megabytes):

DE	RAM Used	RAM Shared	Buffers	Disk Read*
KDE Xorg	1511	45	536	904
KDE Wayland	1487	71	454	910
Gnome Wayland	1422	12	341	834
Gnome Xorg	1355	31	364	846
Cinnamon	858	46	328	785
Mate Compiz	624	17	226	699
XFCE	597	13	224	732
LXQT	491	28	250	739
IceWM (baseline)	271	1	189	-

* - the amount of data read starting from a cold boot to a desktop environment, including a file mananager and graphical terminal emulator.

RAM Used includes all the default background applications, services and daemons which a particular Fedora spin is running.

Whoever doubted XFCE as a lean desktop environment is I hope dissuaded and changed their mind.

Overall if you are a user of KDE and Gnome you must be looking at the very least 4GB of RAM though modern web browser eat RAM for breakfast so 6GB or even 8GB of RAM sound a lot more desirable..

This doesn't sound like a lot of work but I've spent a full day downloading ISO images, installing and updating spins, collecting the data and presenting it. Donatations are impossible considering I live in Russia but at least consider Google Ads on this page. Thank you.

If you need raw data you could download the source image files (zip archive, ~1MB).

P.S. KDE under Wayland was very unstable. On every second boot I couldn't launch any applications after logging in. This could be an issue with VirtualBox though. Gnome worked flawlessly.

P.P.S. A similar comparison for FreeBSD with very similar results.