Intel Linux Optimizations Help AMD EPYC "Genoa" Improve Scaling To 384 Threads

Written by Michael Larabel in Software on 6 April 2023 at 04:00 PM EDT. Page 3 of 5. 13 Comments.
OpenVKL benchmark with settings of Benchmark: vklBenchmark ISPC. CentOS Stream: 384 Threads was the fastest.

As expected and as a reference run / smoke test, for not all workloads was there a measurable difference between the three tested Linux operating systems on the AMD EPYC server.

GROMACS benchmark with settings of Implementation: MPI CPU, Input: water_GMX50_bare. Clear Linux: 384 Threads was the fastest.

With the GROMACS 2023 software package, the results were quite similar until 48 cores/threads and above where this time CentOS Stream was falling behind Clear Linux and Ubuntu.

MariaDB benchmark with settings of Clients: 2048. CentOS Stream: 96 Threads was the fastest.

With the MariaDB 11 MySQL database server, there wasn't much of a difference in the results until running in the full EPYC 9654 2P configuration of 192 cores / 384 threads... In that configuration with SMT active, the Ubuntu 23.04 performance tanked completely while CentOS Stream 9 was flat compared to its 192 core configuration with SMT disabled and Intel's Linux distribution enjoyed a slight uplift in its results.

MariaDB benchmark with settings of Clients: 4096. CentOS Stream: 96 Threads was the fastest.

When pushing the AMD EPYC Zen 4 server harder with 4096 concurrent clients hitting the MariaDB 11 server, CentOS Stream surprisingly was providing noticeably better performance at 24 to 96 cores/threads... Possibly due to its XFS file-system default? When moving from 196 cores with SMT off to SMT on, the Ubuntu Linux performance again tanked, CentOS Stream remained flat, and Clear Linux saw a slight boost to its performance.

MariaDB benchmark with settings of Clients: 8192. Clear Linux: 384 Threads was the fastest.

Lastly for the MariaDB/MySQL testing when loading up the server in the heaviest configuration, CentOS Stream was enjoying nice performance in the 24~96 core range while Clear Linux was scaling better at 96 cores/threads and greater.


Related Articles