Submitted by All the remixes use less memory than the default GNOME edition. To be honest, we didn’t expect that. The last time we did this comparison, in 2013, Kubuntu scoffed the most RAM – and as before, it still uses the most disk. KDE Plasma 5 really has slimmed down its memory footprint impressively, although it’s still no lightweight.
The KDE, MATE, and Budgie editions are quite close in resource usage so in those terms, there’s not a lot to choose between them. That means it’s down to your personal preferences.
All credit to the Lubuntu team: their remix remains the lightest by quite some margin, both in memory and disk usage. Saying that, it does use an old version of the LXQt desktop. There is a repository to install a newer version, but that’s a big ask for a non-techie user.
These differences seem minute and insignificant to me, especially once you start loading a browser with a few tabs or a few documents, and any of these small RAM differences will melt like snow in the Sahara.
Memory footprint affects more than main memory. It affects the caches as well. If code is compact it fits better into the smaller and faster caches. Ideally, the code would fit into the level one code cache in its entirety.
Game devs are well aware of cache loading optimizations and if the main loop of a game doesn’t fit in the L1 cache, the performance nosedives. Comparing the seL4 microkernel to Hurd yields a similar comparison.
SamuraiCrow,
Yeah, so many devs stopped caring about this stuff, but it can make a big difference. It’s not just caches, but TLB & page tables too. Asset space efficiency not only effects run time but load times and task switching time as well due to page evictions.
There can be good reasons to upgrade, but when software becomes less efficient it is kind of one step forward one step backward.
Few companies really want to budget anything for optimization when a project gets “good enough”. Once it works the bosses won’t appreciate spending more money to run smoother on older hardware. Who cares? If users have to buy new hardware then so be it. And thus many hardware advancements are negated by the software industry.
What do you mean exactly by “the main loop?”
L1 nowadays has been tuned to fit most compute kernels that are likely to be encountered in the wild.
I am not arguing for/against using availble memory. But why is gnome so damn slow? Launching the same application in kde or amiwm, icewm or awesomewm shows this problem replicably and nothing is done to fix it. Wayland is a slow turd (at this point) but it is getting better, perhaps in a future with 5ghz 32 core systems we can have the UI speed of BeOS on a dual 233mhz system as shown in the infamous “BeOS demo video”, One can hope,
Read the post before yours about cache loading optimization aka memory hierarchy optimization.
Does that apply even to desktop applications? And does that explain the slow draw rate of gnome compared to other desktops? Why does it take over 20 seconds to launch firefox in gnome whilst in AmiWM and IceWM it takes no time at all?
Tested on the same amd 5950x machine over the entire summer, but added two intel systems into the mix. the 44core intel xeon performed even worse. The laptop 4950 haswell cpu performed well though in gnome and lowered the launch speed by some metrics in gnome.
NaGERST,
Sounds like you need some microbenchmarks to get those answers! I don’t have any machines running gnome, otherwise I might test it.
I’d start by comparing the disk activity, and then I’d try and find other sources of overhead. Is gnome slower because it’s doing more “work”, or is it blocking on excessively chatty IPC? Is it making inefficient use of syscalls? Do the APIs themselves create drastically different overhead? Obviously it’s something, but it could take a lot of testing to get an answer.
I’m in agreement that the game optimization likely has little to do with desktop app performance. I have never noticed gnome being slow even on 5-7 year old workstations with normal windows 10 sized memory (8 GB min) . I wonder how much things like their silly animations are impacting the implied speed. Like it seems slower because it decides to run a dumb animation slowly so you can see it, kind of thing that windows used to do.
In any case gnome works. I’ve never had any stability issues with it, ever. Can’t say the same for KDE ( though I love its UX better) and the lxde is just missing features I kind of need sometimes. But any desktop of mine is required to run wayland thats just not optional.
GNOME is not that “slow.” It mostly out of the way anyways, as it really doesn’t do much. You just wanted to talk about BeOS, me thinks.
also in windows it is generally a bad idea to hog memory, (got a lot better with nt) Even windows 11 has problems with releasing memory occupied by “defined memory pograms” Pink panther games are the prime example, luckily you can play some of them to at least 40% in scummvm. Damn kids wanting to know what happened after spending hours (weeks for them),,
The pink panther games are just about the only good games translated into swedish with voice actors, so they are not just important from a linguistic standpoint, but also a cultural one.
Pink panther games? Thats so interesting… Do you have more information about them? Where do you find them? Not a sweedish speaker, but those seem like they would make games I’d be interested in playing.
They are animated point and click adventure games, good 2d graphics. Something aching in quality to Curse of monkey island. Story is fun and puzzles as well as the animation is well done. The first game is called “Resa på egen risk” or “Rosa resa” depending one which of the publishers they used and is about traveling the world and learning stuff about different cultures, the music acts are pretty nice as well. The second one is less well written, but with the same level of good animation and graphics the 2nd game is named “Hokus pokus panther” and the story is about pink panther finding a magic wand and doing some detective work.
I do not know the titles in english, as i only ever owned and played the swedish ones.
Yeah, “bloatware” is a moving target but I think Gnome definitely qualifies these days. It’s also just poorly engineered on Wayland compared to KDE, IMO.
I’m a bit surprised about how much of a hog the ostensibly lightweight Budgie is, but maybe shouldn’t be – doesn’t it also use Mutter? And unlike Gnome it does not even have Wayland support to show for that yet, last I checked. Too bad, it is a very nice looking desktop.
BTW I posted this story as well as wrote it — and here is the sequel:
https://www.theregister.com/2022/08/19/less_mainstream_xbuntu/
We look at the official flavors, but there are more out there
A very important point that it’s clear from the comments that I should have mentioned: another consideration here is which desktops really benefit from hardware 3D acceleration.
Always uses 3D rendering: GNOME, Cinnamon, Unity, (possibly) Budgie as it’s based on GNOME 3 tech.
(So *none* of these work great in VMs, and ideally you want Intel or AMD 3D GPUs, or the binary nVidia driver, rather than Nouveau.)
Use is optional: works without it, but will use it if it’s there: KDE
Use is not needed: works fine without it, but can use an additional compositor app for fancy effects if you have one: Xfce, MATE, LXQt
Still not sure: UKUI.
But did you check if desktop really benefit from 3d or just used information from comments? What are real benefits, does they make any difference? And how you checked if they do? Are you sure they do? I don`t understand things like “possibly” and “not sure”.
Marshal Jim Raynor,
I’d like to see benchmarks on this too.
Many of the ARM SBCs I’ve tried have laggy feeling on the desktop, and I always assumed it had to do with the bad state of linux GPU acceleration there, but I wasn’t sure if that was the real cause.