With the in-development [1]Linux 6.13 kernel one of the biggest features for those using new AMD EPYC 9005 "Turin" processors is [2]using the AMD P-State driver by default for servers/motherboards with ACPI CPPC support enabled. But even for platforms without that where ACPI CPUFreq remains the default, the Linux 6.13 kernel is still showing some nice incremental uplift at large on these new AMD server processors. Here are some Linux 6.11 vs. 6.12 vs. 6.13 Git kernel benchmarks using an AMD EPYC 9575F 64-core server.

[3]

With the recently assembled AMD EPYC Turin 1P test server built around the [4]Supermicro H13SSL-N motherboard, [5]12 channel DDR5-6000 memory with [6]active memory cooling , and [7]SilverStone XE360-SP5 liquid cooling , I ran some benchmarks of the three latest kernel series for seeing how Linux 6.13 is looking. The Linux 6.13 kernel merge window only wrapped up a little more than one week ago and this kernel won't be out as stable until the end of January, but nevertheless benchmarking curiosity!

With this Supermicro H13SSL-N motherboard though it isn't working properly with the AMD P-State driver due to ACPI CPPC issues being investigated and as a result continues using the ACPI CPUFreq driver even on Linux 6.13. So while Linux 6.13 is exciting for amd_pstate on EPYC 9005 series by default for supported configurations, in the benchmarks today this article is using ACPI CPUFreq across all three kernel versions. In any event the Linux 6.13 benchmarks are looking promising. From the Volcano reference server I did run some initial AMD P-State benchmarks, those interested in that ACPI CPUFreq vs. AMD P-State comparison see Early Benchmarks: [8]AMD EPYC 9005 Performance & Power Efficiency To Lead Further With Linux 6.13 .

[9]

There are [10]many exciting features and changes in Linux 6.13 and I've just begun my testing of Linux 6.13 Git on different desktops and servers. I'll have Linux 6.13 kernel benchmarks on more systems soon but in any event what I am seeing so far from this EPYC 9575F server is quite enticing even with the same CPU frequency scaling driver at play. Testing on the additional systems will also help show the root cause of these Linux 6.13 performance changes and how widespread they may be.

The same kernel configuration was used across all the upstream Linux v6.11, v6.12, and v6.13 Git kernel builds. No other software or hardware changes made during testing besides rebooting and swapping out the kernel in use each time.



[1] https://www.phoronix.com/search/Linux+6.13

[2] https://www.phoronix.com/news/Linux-6.13-Power-Management

[3] https://www.phoronix.com/image-viewer.php?id=linux-613-amd-epyc-9575f&image=linux_613_epyc_1_lrg

[4] https://www.phoronix.com/news/Linux-6.13-Power-Management

[5] https://www.phoronix.com/review/8-12-channel-epyc-9005

[6] https://www.phoronix.com/review/corsair-ddr5-ram-cooling-server

[7] https://www.phoronix.com/review/silverstone-epyc-sp5

[8] https://www.phoronix.com/review/amd-epyc-9005-pstate

[9] https://www.phoronix.com/image-viewer.php?id=linux-613-amd-epyc-9575f&image=linux_613_epyc_2_lrg

[10] https://www.phoronix.com/review/linux-613-features