Linux 5.0 Through Linux 5.4 Benchmarks On AMD EPYC 7642 "Rome" Server

A month ago I posted benchmarks looking at the performance of Linux 4.16 through Linux 5.4 kernels using an Intel Core i9 workstation. Stemming from that was a request for an AMD EPYC kernel comparison, so I carried out said tests. Due to the Rome support being newer, this round of testing is looking at the EPYC 7642 performance on Linux 5.0 to Linux 5.4.

The tests were done last month but with the results not being too interesting, publishing them escaped my mind until this week firing up some of the Linux 5.5 kernel benchmarks. Those initial Linux 5.5 numbers for AMD EPYC should be out in the days ahead and should be interesting given some prominent changes with Linux 5.5 and other early performance numbers showing some interesting changes.

For this testing Linux 5.0 through Linux 5.4 Git was tested on the same EPYC Rome server.

But without any recent vulnerabilities and mitigations like was the case with Intel as recently as their November disclosures, the Linux 5.0 through Linux 5.4 performance with EPYC Rome wasn't all that interesting and mostly just the usual movement/noise we see in some select tests.

The geometric mean of all the test results basically point to flat performance for Linux 5.0 through 5.4, which is better than the performance hits seen in the recent Intel Linux kernel comparison.




A few real-world tests were trending up with Linux 5.3~5.4.


And some small kernel scheduler hits since Linux 5.3, but with 5.5 and the scheduler changes there the Hackbench numbers should be more interesting.

While not all of the same tests and other system differences, here was the i9 workstation results on Linux 4.16 through 5.4 for reference.

But for the most part of 76 tests ran, there wasn't much in the way of surprising movement for this EPYC testing of 5.0 through 5.4. The rest of the data can be found via this OpenBenchmarking.org result file. The upcoming Linux 5.5 results should be more interesting.