Linux 6.6 Looks To Be Very Lucrative For AMD Server Performance

As a continuation of last week's article looking at Linux 6.6 bringing some impressive gains for AMD EPYC Bergamo, over the past few days I've also tested Linux 6.5 stable and Linux 6.6 Git on Genoa and Genoa-X processors as well as Intel Xeon Scalable "Sapphire Rapids" in looking at this next kernel version's performance. The Sapphire Rapids performance was largely flat while for an interesting class of workloads the Linux 6.6 performance drives the AMD EPYC server performance much higher.

Last week's article showed that on the AMD EPYC 9754 a variety of database servers, TensorFlow, and other interesting workloads were exhibiting some very nice gains with Linux 6.6 over Linux 6.5 stable. The Linux 6.6 kernel brings the EEVDF scheduler, workqueue improvements to benefit chiplet-based CPUs with multiple L3 caches like AMD's wares, and much more. There are many new features with Linux 6.6 and AMD server CPU performance seems to be benefiting nicely. In my testing of AMD Ryzen CPUs with Linux 6.6, it's largely been flat for those Linux desktop workloads.

Today's Linux 6.5 stable vs. Linux 6.6-rc1 benchmarking is looking at the 1P performance of the EPYC 9754 Bergamo, EPYC 9684X Genoa-X, and EPYC 9654 Genoa processors alongside a Xeon Platinum 8490H Sapphire Rapids processor. Ubuntu 23.10 in its current development state was run across all the tested platforms. Linux 6.5.1 and 6.6-rc1 were obtained from the Ubuntu Mainline Kernel PPA for easy reproducibility and analysis by third parties. No other changes were made besides switching out the kernel version in use between runs on each Intel/AMD processor.