Fedora, Ubuntu, CentOS, openSUSE, Debian, Clear & Antergos Linux Benchmarks On AMD EPYC

[phoronix]

Phoronix: Fedora, Ubuntu, CentOS, openSUSE, Debian, Clear & Antergos Linux Benchmarks On AMD EPYC

I've just wrapped up trying out nine different Linux distributions on AMD's EPYC in the form of the EPYC 7601 housed in the TYAN Transport SX TN70A-B8026. Like our initial testing with Ubuntu on EPYC, the other modern Linux distributions all played nicely with AMD's re-entry into the server market with their Zen-based offerings. But as with any new CPU platform, the out-of-the-box performance can vary greatly depending upon the Linux operating system being used. Here are benchmarks including Fedora, Ubuntu, CentOS, openSUSE, Debian, Clear Linux and Antergos.

http://www.phoronix.com/vr.php?view=25294

[tomtomme]

@Michael: The "Wins" vs. "Losses" section is nice, but could be improved in its representativeness.
You would need to add up the ranks of each distro and divide by the number of benchmarks. For example: If Fedora ranked 1., 3., 4., 2., (=10) in 4 benchmarks the mean rank would be 2.5.
Problem could be to actually get a rank. You would need to take into account if the benchmark is a "the lower the better" or "the higher the better"

[Michael]

@Michael: The "Wins" vs. "Losses" section is nice, but could be improved in its representativeness.
You would need to add up the ranks of each distro and divide by the number of benchmarks. For example: If Fedora ranked 1., 3., 4., 2., (=10) in 4 benchmarks the mean rank would be 2.5.
Problem could be to actually get a rank. You would need to take into account if the benchmark is a "the lower the better" or "the higher the better"

Getting HIB/LIB isn't an issue. If anyone wants to contribute improvements, that part of the code is really easy:

https://github.com/phoronix-test-suite/phoronix-test-suite/blob/master/pts-core/commands/winners_and_losers.php

[eva2000]

I'm curious what were Redis's specific GCC compiler paramters for Intel Clear Linux ? The chart sort of cuts off the right of the options used.

[elldekaa]

I'm curious what were Redis's specific GCC compiler paramters for Intel Clear Linux ? The chart sort of cuts off the right of the options used.

You can select it to see them:

-pipe -fexceptions -fstack-protector -m64 -ffat-lto-objects -fno-signed-zeros -fno-trapping-math -fassociative-math

[andyprough]

I'd be interested to see how Tumbleweed would perform if EXT4 was used instead of hamstringing it with Btrfs and XFS.

Clear Linux holds zero interest for me, since it's not even a real distro that can do the work of the others. It's just a concept distro for cloud containers or for benchmarking. I could never get enough software pieces pulled together to do any real work with it.

[tchiwam]

Is it possible to make a matrix with kernel and userland to know which improvement belongs to the kernel compile config and what belong to the binary and library ?

[Zoll]

So why aren't Debian based distros using "aggressive compiler" optimizations such as Clear Linux? It's obviously doing something right.

[tomtomme]

Getting HIB/LIB isn't an issue. If anyone wants to contribute improvements, that part of the code is really easy:

https://github.com/phoronix-test-sui...and_losers.php

sadly I am not really a coder. I did some html 10 years ago, thats it. I can read the code sure, but if I edit it - how would I even test the changes?

[eva2000]

You can select it to see them:

-pipe -fexceptions -fstack-protector -m64 -ffat-lto-objects -fno-signed-zeros -fno-trapping-math -fassociative-math

thanks !