Fedora 38 Beta Performance Mostly Flat, Few Regressions

[phoronix]

Phoronix: Fedora 38 Beta Performance Mostly Flat, Few Regressions

For those curious how the performance of Fedora 38 is looking ahead of its official release at the end of April, here are some preliminary benchmarks looking at the performance of this leading-edge Linux distribution as of the Fedora 38 Beta milestone last week. On both Intel Core i9 13900K "Raptor Lake" and AMD Ryzen 9 7950X "Zen 4" desktop systems, the Fedora 37 performance was compared to that of Fedora 38 Beta.

https://www.phoronix.com/review/fedora-38-beta-benchmarks

[ehansin]

Cannot comment on performance, but I updated my Fedora 37 Workstation (so standard desktop) installation to 38 Beta. I kept want to do some sort of Arch minimalist or for that matter Fedora minimalist installation. I've learned a lot over the years fiddling with different distros, lots of various server installs, etc. But I needed to stop fiddling and get to work. I wanted a nice development environment from my Linux boot. Finally just went ahead with Fedora Workstation when 37 released. Gnome looked so much nicer than the 3.38 I last remembered, I just let it install the core "infrastructure" vs. me trying to build up, Wayland native, I have Sway installed as an login option, etc. So performance aside as performant enough for me, all in all a great release.

[user1]

Is -fno-omit-frame-pointer already enabled in 38 beta? If yes, then I wonder if that's the reason for some of the regressions

[fagnerln]

wow... 20 to 30% regressions on AMD CPU is too much, no? Someone has idea of what's happening?

[EDIT] I have a pretty low end AMD CPU (Athlon 200), if I lose performance in games, I'll need to jump to other distro.

[AdamW]

wow... 20 to 30% regressions on AMD CPU is too much, no? Someone has idea of what's happening?

[EDIT] I have a pretty low end AMD CPU (Athlon 200), if I lose performance in games, I'll need to jump to other distro.

well, it's fairly likely the problem would show up in other distros eventually too if it's not identified; we don't change much about the kernel/graphics stack downstream in Fedora compared to upstream, so it's likely this is caused by some kinda issue in upstream kernel or mesa or something. It's an interesting finding, though I've no idea at all what the cause might be. I'll ask our kernel/gfx folks to take a look.

[avis]

"The updated Fedora 37 install and Fedora 38 Beta releases"

mostly feature the same packages versions except glibc and gcc which are normally only updated with a new release. It doesn't make a lot of sense to compare the two.

[You-]

"The updated Fedora 37 install and Fedora 38 Beta releases"

mostly feature the same packages versions except glibc and gcc which are normally only updated with a new release does it doesn't make a lot of sense to compare the two.

Different compiler flags. There were some [I am exaggerating here] "woe to us, the world will end" type suggestions and debates about the performance impact of enabling a couple of them. It seems they didnt make any different to performance.

[Anvil]

just wonder if its a bit of both, kernel an Graphix Stack

[GraysonPeddie]

I would definitely want to see some more gaming benchmarks for Fedora 38 Beta, but then I wonder if this will affect unreleased Ubuntu 23.04 as well...

[Britoid]

wow... 20 to 30% regressions on AMD CPU is too much, no? Someone has idea of what's happening?

[EDIT] I have a pretty low end AMD CPU (Athlon 200), if I lose performance in games, I'll need to jump to other distro.

If you're on Fedora you're probably better off running your games in the various Flatpaks (e.g Steam, Lutris etc) anyway which don't afaik have omit-frame-pointer.

It would actually be cool to see Phoenix also run the tests inside the Flatpak env, would make it more easy to see regressions in the kernel.