Nick of The Linux Experiment posted a comparison of resource usage for Desktop Environments, that could be quite useful, particularly when it comes to tailoring options to older hardware.
Below is a summary of sorts. RAM seems to be the real standout, with a nearly 100% difference between lowest usage and highest usage.
All tests were performed on a Slimbook Evo 14, with Fedora 42 beta as the base OS.
Tested desktop environments include:
Budgie
Cinnamon
Cosmic
Gnome
KDE Plasma
LXQt
XFCE
Ram Usage MiB (total 15,260 MiB)
1,310 MiB / +96% difference between XFCE and KDE Plasma
XFCE = 1,360
LXQt = 1,370
Budgie = 1,450
Cinnamon = 1,890
Cosmic = 1,970
Gnome = 2,100
KDE Plasma = 2,670
CPU Performance | Single Core (Geekbench 6)
106 points / 4% difference between Cosmic and Gnome
Cosmic = 2,713
LXQt = 2,706
XFCE = 2,700
KDE Plasma = 2,695
Budgie = 2,691
Cinnamon = 2,648
Gnome = 2,607
CPU Performance | Multi Core (Geekbench 6)
299 points / 2.7% difference between Budgie and Gnome
Budgie = 11,499
Cosmic = 11,498
LXQt = 11,496
KDE Plasma = 11,450
Cinnamon = 11,417
XFCE = 11,366
Gnome = 11,200
GPU Performance | 1080p (Unigine Heaven)
20 points / 2.4% difference between Cosmic and LXQt
Cosmic (Wayland) = 854 / 33.9fps avg
Budgie (X11) = 850 / 33.7fps avg
XFCE (X11) = 848 / 33.7fps avg
KDE Plasma (Wayland) = 846 / 33fps avg
Cinnamon (X11) = 845 / 33.6fps avg
Gnome (Wayland) = 844 / 31fps avg
LXQt (X11) = 834 / 33.1fps avg
Battery Life usage % per hour (Firefox video loop)
All results fall into just two outcomes 50% usage difference between group #1 and #2
IMO, Fedora is a terrible choice for that testing. Their spins are very different as far as what is installed and running by default. The Fedora KDE spin is very…uh…fully featured.
Those numbers also contradict what we have seen reported by our users in the past as well.
Yeah I wondered at the use of Fedora. My bias is obviously towards Arch because at least everything is running the latest possible and bare minimum.
The other thing missing from these tests (and synthetic benchmarks generally) is desktop environment responsiveness, which is really difficult to objectively measure. Things like opening and closing applications, creating tabs in browses and loading pages, etc. On old hardware, some of these mundane everyday tasks become tedious and in the hungrier desktop environments, such as KDE Plasma for example.
The RAM comparison is perhaps the only current reflection of that. The other tests show negligible difference in CPU / GPU, so they don’t really touch on that element of user experience.
I’ve never had that sort of RAM usage on Arch and GNOME.
At fresh boot for one system with 8G memory, I have around 850-900 MB memory usage.
On another one with 16G, I have around 1100-1200 MB.
I haven’t done any benchmarking regarding performance or anything else.
Though, one thing that I have noted is battery life. On the one with 16G (Lenovo Yoga with AMD Ryzen 7 4800U with Radeon Graphics) I get considerably better battery life on GNOME than on Plasma which is installed in another Btrfs subvolume.
The KDE Plasma system uses also more memory. Around 1350-1400 MB.
All these system were quite minimally setup at the outset, but they have gathered a bit of cruft over time.
I have done a bunch of RAM tests with KDE plasma and XFCE and plasma has always used slightly less memory although it isn’t different enough to matter much. Like ~100MB difference. XFCE was lightweight when it was gtk2-based but once them moved to gtk3, memory usage went up by a lot.
IMO, the biggest differences are two things. Plasma has many effects on by default, if you have a very old integrated CPU, turning those off will make a big difference and make the desktop feel much more responsive.
The other thing is that plasma does a lot of caching. If you are using an HDD or some other form of slower storage, that will drag plasma to a crawl. You really want an SSD for plasma.
I’m feeling inspired to design and conduct some comparative benchmarks for EndeavourOS, on modest 10 year old hardware. It’s in the extremes that the differences become more obvious, and older hardware can demonstrate that. It’s just a matter of finding some time for it.
I thought I had disabled all the effects under KDE Plasma on that old system, but it’s been a while, so wouldn’t hurt to try again.
The perceived performance difference between KDE Plasma and LXQt was night and day.
The system in question has a single Kingston KC3000 NVMe 512GB in it, so storage speed shouldn’t be an issue.
If I had the interest and the energy I would do this type of test with my own Arch Linux based ISOs. I spin up seven desktops: Cinnamon, Gnome, KDE, Lxde, Lxqt, Mate, and Xfce. My Ezarcher spins each contain the LTS kernel, so maybe the interest would not be so high? Other than the DE’s, my ISOs are all the same underneath the desktops.
Except for the first test, which was overall ram usage, the balance of tests were neglibile in this layman’s eyes.
AFAIK Budgie has always been that way with laptop batteries. I thought the Fedora Budgie was meh. Endeavour Budgie much better. Solus the best. But some weirdo Mandriva spin was far out,
As far as Cinnamon, Endeavour is the best experience I ever had for Cinnamon.
Plasma? See: baloo.
Like Dalto said, Spins are an odd choice the TilVids site decided but there are distros that have 100% fan-curated Spins with little input from the company. I think Fedora takes an active role in the Spins.
The absolute flat out consistency of the kingston ssd, that particular hardware, and the Evo is a variable all to itself as well.
Really interesting reading.
Even the indexing has been tamed a lot in more recent versions… a lot less people complaining about baloo (whether its finished its indexing or not).
I agree that for the most part XFCE cant quite be considered ‘light’ like it once was .. and Plasma similarly doesnt really deserve to be considered extra ‘heavy’.
If one wants the nimblest bare bones they will need to use less than a full DE - such as a WM.
(Both the assorted Window Managers and GNOME can be considered to be of separate classes.)
I like Nick for his general Linux news coverage but like he often says himself:“Take this with a grain of salt”. While i appreciate the effort and results for Fedora 42 by him, i would love to see a more tech oriented Linux streamer like Brodie do the same experiment for Arch and perhaps Chris Titus Tech for Debian.
Something like that could give us a more broad perspective on the subject. My 2 cents!
I seem to remember reading somewhere that if there’s enough ram, Plasma will use more to improve efficiency and speed, whereas for systems with small amounts of ram it uses less.
I can’t remember the thresholds but I do know that it runs on a 2G machine with about 400-600M free,
RAM usage it seems, is perhaps not the most useful metric for how heavy a desktop environment is.
I’m keen to try and quantify real-world usage experiences, in a measurable and repeatable way. I can subjectively say LXQt feels much snappier than KDE Plasma on slow old hardware. But that’s subjective… so… so what.
Objective, measurable, repeatable tests that capture real-world usage, would be valuable in helping decide the right fit of desktop environment. Hmm… something to think on.
And I’m gonna say no one size fits all, so it’s probably going to end up being, ‘Try it, you might like it’ vs a scientific quantifiable recommendation.
I’m just grateful the Linux world has choices.
