Intel Continues To Show AMD The Importance Of Software Optimizations: 16% More Ryzen 9 9950X Performance

[phoronix]

Phoronix: Intel Continues To Show AMD The Importance Of Software Optimizations: 16% More Ryzen 9 9950X Performance

As part of my ongoing AMD Ryzen 9 9950X Linux testing, last week I provided a look at the AVX-512 benefits to Zen 5 and also the Windows vs. Linux performance for the Ryzen 9 9950X. For sharing today is a look at multiple Linux distributions up and running on the AMD Ryzen 9 9950X (Zen 5) desktop. Among the distributions in the mix are Intel's Clear Linux distribution that is optimally tuned for maximum x86_64 Linux performance and once again even on AMD hardware shows the significant benefits to a well-tuned Linux software stack.

https://www.phoronix.com/review/linux-os-amd-ryzen9-9950x

[arunbupathy]

Man! What is Fedora doing wrong that it's dead last in a lot of these benchmarks, even compared to Ubuntu?

[Anux]

If Cachy uses that much more optimizations it somehow doesn't pay off, also many regressions. :/

[V1tol]

CachyOS appears to be a great distro - all the Arch benefits combined with great performance. Clear is obviously better but it doesn't look like a daily driving distro for me.

[skeevy420]

Michael
On CachyOS did you use the x86-64-v4 or Zen 4 repos? A new enough install should have used Zen4 but the screenshot is using x86-64-v4.

[ptr1337]

Michael
On CachyOS did you use the x86-64-v4 or Zen 4 repos? A new enough install should have used Zen4 but the screenshot is using x86-64-v4.

Its used the Zen4 repo

Thanks for the benchmarks, Michael!
Mainly in C-Ray, Clickhouse and the Tensorflow/Pytorch.

Even tough, lets take the "Liquid-DSP", "Rocksdb" Benchmarks, which seems to be compiled from the Benchmark Suite, which (once again) does not respect the CFLAGS from the /etc/makepkg.conf.
I really dont want to put them into /etc/environment, just to compare with clearlinux there.
Michael , would it be possible to adjust PBS to also read /etc/makepkg.conf, or simply ignore this setting for any distro? This would make such distro benchmark in a way better comparision.

[ptr1337]

If Cachy uses that much more optimizations it somehow doesn't pay off, also many regressions. :/

You never can avoid regressions. Many libs used their, were not even using the compiled repository packages. Some were compiled by PBS, some were just binaries.
There could be also regressions, in the used BORE Scheduler, which we might not found yet or the 6.11 amd-pstate patches.

We want to setup soon, some kind of benchmark server, which does on a weekly/monthly basis benchmarks, compared to archlinux with the same versions as well as against our own packages previously.
Donations are raising, but we preferred to cover the infrastructure better first (Mirrors, Zen4 repo, Storage Server), before getting just another expensive dedicated server.

[varikonniemi]

Why is no-one able to pin down what clear linux does that is so much better than the rest? Even cachyos that is supposedly optimized was always slightly behind, except when it regressed badly, or or failed to leap ahead like clear.

Clear has been for a long time in a class of it's own, and it's embarrassing others have not been able to learn from it, we are talking open source after all.

[Jonjolt]

Man! What is Fedora doing wrong that it's dead last in a lot of these benchmarks, even compared to Ubuntu?

Did some of there proposed changes to improve debugging go through?

Also SELinux, for what it does you'd think it would be slower

Though I wonder why it does great with Java /s

[skeevy420]

What's up with the comments lately? Ever since the Phoronix UI updated it seems like it can take upwards of 20 to 30 minutes for people's posts to appear even though I've been given email notifications about being quoted.