Early Benchmarks: AMD EPYC 9005 Performance & Power Efficiency To Lead Further With Linux 6.13

Written by Michael Larabel in Processors on 12 November 2024 at 11:10 AM EST. Page 3 of 4. 1 Comment.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Slow. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Slow. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Slow. acpi-cpufreq performance was the fastest.

Video encoding is one of the workloads where the CPU frequency driver/governor can make a significant difference to performance especially on high core count servers. With H.265 video encoding of 4K content with Kvazaar, the amd-pstate-epp powersave configuration is yielding much better performance than the acpi-cpufreq schedutil default currently used on Ubuntu 24.04 LTS. The amd-pstate-epp powersave configuration was yielding the best performance-per-Watt of the tested configurations.

Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Very Fast. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Very Fast. acpi-cpufreq performance was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Very Fast. acpi-cpufreq performance was the fastest.

Across a variety of configurations with the Kvazaar H.264 4K video encoding, the amd-pstate-epp powersave configuration as the default for Linux 6.13+ was showing the best performance-per-Watt and much better than acpi-cpufreq schedutil.

Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. amd-pstate-epp performance was the fastest.
Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. amd-pstate-epp performance was the fastest.

When compiling the Linux kernel with all modules in an x86_64 kernel build, the amd-pstate-epp powersave build time was shorter than acpi-cpufreq schedutil and similar to the performance governor runs. Beyond amd-pstate-epp powersave being faster than acpi-cpufreq, making it all the more exciting is that the combined AMD EPYC 9755 power consumption was around 35 Watts lower on average than acpi-cpufreq or 20~30 Watts lower than the performance governor runs.

Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. amd-pstate-epp performance was the fastest.
Timed Linux Kernel Compilation benchmark with settings of Build: allmodconfig. amd-pstate-epp performance was the fastest.

The slightly lower power use can have some minor benefit on the CPU thermal side.

Timed LLVM Compilation benchmark with settings of Build System: Ninja. amd-pstate-epp performance was the fastest.
Timed LLVM Compilation benchmark with settings of Build System: Ninja. amd-pstate-epp performance was the fastest.
Timed LLVM Compilation benchmark with settings of Build System: Ninja. amd-pstate-epp performance was the fastest.
Timed LLVM Compilation benchmark with settings of Build System: Ninja. amd-pstate-epp performance was the fastest.

The build times for LLVM with the Ninja build system were similar across the tested configurations, but the combined EPYC 9755 2P power consumption with amd-pstate-epp powersave was around 30 Watts lower than the rest.

Related Articles