CachyOS Had A Really Great Year Advancing This Performance-Optimized Arch Linux Platform

[MadCatX]

I do not understand your point. The test says: "All the Linux distributions were cleanly installed and tested in their default / out-of-the-box state except for the noted extra runs on Ubuntu for reference purposes."

So cachyos binaries where tested as provided by the distro. This has nothing to do with /etc/makepkg.conf which is only used if you compile packages by yourself.

I don't think that's quite true. AFAIK PTS builds the programs in then benchmarks from scratch to ensure maximum reproducibility across different systems and different benchmark runs. It needs to do this mainly to make sure that it uses the same version of the benchmarked program every time. Unfortunately, if PTS or the build system cannot pick up any additional compiler settings, it will use whatever the default is for the benchmarked program. You can see the effect of this in benchmarks where Clear Linux pulls ahead of the competition because it exports its additional compiler flags through /etc/environment. CachyOS's narrow victory probably goes on the account of the low-level binaries (glibc, libgmp, etc..) being built with optimized CFLAGS.

A much more telling benchmark would be just Arch vs. CachyOS using only distro-packaged binaries. Assuming that CachyOS is not too far behind Arch in terms of package updates, such a benchmark shouldn't be too difficult to set up...?

Michael: Any chance you could shed some light on this? Also, Merry Xmas

[isaacx123]

The article says: "From the performance side, the CachyOS optimizations are beginning to pay off."

I do not agree with this statement. Yes, cachyos is leading in this benchmark. But Arch Linux is second and cachyos is in average just 1 % faster than Arch Linux (2284 vs. 2253 points). That tells me that all the performance tweaks in cachyos are NOT paying off.

Well, at least when it comes to gaming, I have seen CachyOS pull ahead of distros like Bazzite or Endeavour by like 15fps on average.

[Nozo]

I agree with this statement. I recently experimented with Cachyos and while certainly not a bad distro ......
There was absolutely no reason to leave Arch. Perhaps at some point in future (which I doubt) there may be reason to switch. But certainly not today.
Arch + Zen kernel is all I could ask for.

There's no reason to leave Arch either, you can install the Cachy repos in your existing install, on almost any Arch-based distro (except Manjaro with the stable repos, for obvious reasons).

You can then proceed to install packages such as cachyos-settings and others to get its defaults, the CachyOS team provides support to both CachyOS and Arch Linux users.

[Nozo]

Well, at least when it comes to gaming, I have seen CachyOS pull ahead of distros like Bazzite or Endeavour by like 15fps on average.

Bazzite has the handicap of using Flatpak for everything, which decreases performance especially in games highly dependent on the CPU, so it is not an entirely fair comparison in that case.

[MadCatX]

Well, at least when it comes to gaming, I have seen CachyOS pull ahead of distros like Bazzite or Endeavour by like 15fps on average.

Were those 15 fps 2 % or 20 % more than what the competition did? Also, what games are we talking about? I really don't imagine that CachyOS could have any meaningful advantage in AAA games that are distributed as binaries and usually avoid the use of system libraries for the bits that are critical for performance.

[mobadboy]

Which distro is better for schoolkids on vacation, CachyOS or Kali? 😏

Kali should run in a VM

Cachy is something of a gimmick, although there is an audience for it so more power to them. Afaik the main benefit is the kernel scheduler. The increasingly optimized binaries are mostly placebo.

Bazzite has the handicap of using Flatpak for everything, which decreases performance especially in games highly dependent on the CPU, so it is not an entirely fair comparison in that case.

Not true

It might be a little slower to open but that's where the perf hit stops. It runs the same code on the same CPU.

[mbod]

A much more telling benchmark would be just Arch vs. CachyOS using only distro-packaged binaries.

Binaries like embree, kavazaar, svt-av1, ospray or blender are all provided by cachyos resp. arch repos. I doubt that they are compiled from scratch for the benchmark.

And if a program is compiled for the benchmark, it will still use the libraries and kernel provided by the OS.

[MadCatX]

Binaries like embree, kavazaar, svt-av1, ospray or blender are all provided by cachyos resp. arch repos. I doubt that they are compiled from scratch for the benchmark.

It depends. As a rule of thumb, if you see the compiler options written next to the results chart, PTS compiled the bechmarked program from source. This is true for Kvazaar or SVT-AV1 for instance. Embree or Blender benchmarks download a prebuilt binary but always from the same source regardless of the system the benchmark is run on.

It needs to be done like this because different distros package different versions of the benchmarked programs.

And if a program is compiled for the benchmark, it will still use the libraries and kernel provided by the OS.

Yeah, but both the kernel and essential system libraries like glibc already have tons of hand-crafted optimizations for the bits that really matter. Spicing them up with compiler flags probably doesn't help that much. This might be the reason why CachyOS doesn't seem to be doing any better than Arch in most of the tests.

[Daktyl198]

Binaries like embree, kavazaar, svt-av1, ospray or blender are all provided by cachyos resp. arch repos. I doubt that they are compiled from scratch for the benchmark.

And if a program is compiled for the benchmark, it will still use the libraries and kernel provided by the OS.

You can download the PTS right now and see which packages are installed on an arch-based system. None of the ones you listed are included in the list. PTS mainly just downloads the development libraries/tools required to download (git/svn/curl) and then compile the tests. This is important specifically for cachyos, as the compilation is utilizing their optimized libraries, but NOT using their compiler flags by default. So the PTS results are not the same results you'd get from downloading those applications from the cachyos repos.

[Dukenukemx]

Does CachyOS have a x86-64-v2 repository?