8-Way Linux Distribution Comparison On A Dual Xeon Scalable Gold Server

When running the Redis server on this Tyan 1U server, Clear Linux had a noticeable advantage out-of-the-box against the other tested distributions.

While running the Blender 3D modeling software on the CPU had the best performance with CentOS 7, a.k.a. Enterprise Linux 7.

CentOS 7 had another win with Xsbench.

Lastly is a PHP benchmark. With the testing done out-of-the-box on each OS, with openSUSE Leap and CentOS 7 sticking to PHP5 rather than PHP7, they are much slower than the others. Fedora 26 meanwhile was delivering the fastest out-of-the-box performance.

For those curious about which Linux distribution is the fastest when running on a modern dual socket server, this time in the form of a Tyan GT24E-B7106, I hope you found these numbers insightful. Granted, with any Linux distribution you can easily customize it to your own liking and needs but in the enterprise Linux environment you are not often spinning your own Linux kernel version, compiler release, etc. With this we're just looking at the out-of-the-box offering of each distribution for reproducibility and expecting each software vendor to provide what they believe to be the sane and best defaults.

While there is a wide assortment of results so you can focus on the workloads most relevant to your use-cases, if wanting an overall metric:

WINS:
Clear Linux 17210:   14  [45.2%]
Debian 9.1:          10  [32.3%]
CentOS Linux 7:      4   [12.9%]
Ubuntu 16.04.3 LTS:  1   [3.2%]
openSUSE Leap 42.3:  1   [3.2%]
Fedora 26:           1   [3.2%]

LOSSES: 
Ubuntu 16.04.3 LTS:            12  [38.7%]
openSUSE Tumbleweed 20170823:  5   [16.1%]
Antergos 17.8:                 5   [16.1%]
openSUSE Leap 42.3:            3   [9.7%]
Fedora 26:                     2   [6.5%]
CentOS Linux 7:                2   [6.5%]
Debian 9.1:                    1   [3.2%]
Clear Linux 17210:             1   [3.2%]

But keep in mind that Phoronix Test Suite generated metric is looking specially at the outright winners and losers and not taking into effect the margins by which the distributions won, etc. You can analyze all of these numbers in fine detail via this OpenBenchmarking.org result file generated by the Phoronix Test Suite.