An Early Look At The AMD EPYC Performance With The In-Development Linux 5.4 Kernel
While the Linux 5.4 cycle just officially began last week and its feature merge window not even over until this weekend, given there are AMD EPYC load balancing improvements and many other kernel improvements in general, I was eager to fire up the in-development kernel on the EPYC 7002 "Rome" series to see how the performance is looking.
With the Linux 5.4 Git state as of a few days ago, I ran some preliminary benchmarks of Linux 5.4 at the time compared to Linux 5.3.0 and Linux 5.2.16 on the Rome "Daytona" reference platform with the EPYC 7642 and EPYC 7742 processors in both 1P and 2P configurations.
Via the Phoronix Test Suite a range of benchmarks were done for looking at the Linux 5.4 performance at this very early stage. As the Linux 5.4 release candidates begin, tests on more hardware will obviously come -- these are just some early numbers with being benchmarking junkies.
Aside from swapping out Linux 5.2 vs. 5.3 vs. 5.4 branches as well as the different EPYC 7002 series processors, the hardware/software remained the same throughout testing.
While there is the EPYC load balancing change, with no major power management changes or other big ticket items for Linux 5.4, we weren't expecting any dramatic differences especially with workloads not really interacting directly with the kernel. With tests like NAMD, the performance was basically flat between the tested kernels.
With the highly-threaded renderers like Blender and Tungsten, the Rome 2P processors did appear to be lifted ever so slightly by the Linux 5.4 kernel but within the margin of error and we'll need to test on more CPUs for any level of certainty.
While firing up the Stress-NG kernel micro-benchmarks there was a notable increase in performance on the EPYC 7642/7742 processors in their dual socket configuration.
The Apache web server performance did see some improvement off this early Linux 5.4 kernel snapshot as well sans the EPYC 7642 1P.
That's the brief glimpse of the Linux 5.4 performance on AMD EPYC 7002 series in some of the early testing thus far. Stay tuned for more benchmarking as the Linux 5.4 kernel matures.
With the Linux 5.4 Git state as of a few days ago, I ran some preliminary benchmarks of Linux 5.4 at the time compared to Linux 5.3.0 and Linux 5.2.16 on the Rome "Daytona" reference platform with the EPYC 7642 and EPYC 7742 processors in both 1P and 2P configurations.
Via the Phoronix Test Suite a range of benchmarks were done for looking at the Linux 5.4 performance at this very early stage. As the Linux 5.4 release candidates begin, tests on more hardware will obviously come -- these are just some early numbers with being benchmarking junkies.
Aside from swapping out Linux 5.2 vs. 5.3 vs. 5.4 branches as well as the different EPYC 7002 series processors, the hardware/software remained the same throughout testing.
While there is the EPYC load balancing change, with no major power management changes or other big ticket items for Linux 5.4, we weren't expecting any dramatic differences especially with workloads not really interacting directly with the kernel. With tests like NAMD, the performance was basically flat between the tested kernels.
With the highly-threaded renderers like Blender and Tungsten, the Rome 2P processors did appear to be lifted ever so slightly by the Linux 5.4 kernel but within the margin of error and we'll need to test on more CPUs for any level of certainty.
While firing up the Stress-NG kernel micro-benchmarks there was a notable increase in performance on the EPYC 7642/7742 processors in their dual socket configuration.
The Apache web server performance did see some improvement off this early Linux 5.4 kernel snapshot as well sans the EPYC 7642 1P.
That's the brief glimpse of the Linux 5.4 performance on AMD EPYC 7002 series in some of the early testing thus far. Stay tuned for more benchmarking as the Linux 5.4 kernel matures.
2 Comments