AMD P-State EPP Performance With EPYC On Linux 6.3

Written by Michael Larabel in Software on 3 April 2023 at 11:30 AM EDT. Page 2 of 6. 9 Comments.
NAMD benchmark with settings of ATPase Simulation, 327,506 Atoms. amd_pstate_epp performance performance was the fastest.

First up was NAMD where there was barely any change between all of the tested CPU frequency scaling drivers/governors except for amd_pstate powersave performing much worse than the rest... But that's not exactly a surprise.

NAMD benchmark with settings of ATPase Simulation, 327,506 Atoms. amd_pstate_epp performance performance was the fastest.

In the amd_pstate powersave mode, the EPYC 7773X didn't even clock above 400Mhz in its quest to conserve power... The other configurations were running meanwhile with a peak frequency of 2.6GHz to 3.0GHz.

NAMD benchmark with settings of ATPase Simulation, 327,506 Atoms. amd_pstate_epp performance performance was the fastest.

AMD P-State powersave obviously saves a lot of power if you want to severely handicap your system. For the other configurations the AMD P-State EPP power consumption was within a watt or two of the other performance configurations.

OpenFOAM benchmark with settings of Input: drivaerFastback, Medium Mesh Size, Execution Time. acpi_cpufreq performance was the fastest.

For the OpenFOAM 10 CFD benchmark, the ACPI CPUFreq driver with performance governor managed to take first place still by a very thin margin compared while now at least AMD P-State EPP was about a second (or less than a percent) faster than the existing AMD P-State driver in its ondemand and performance governor modes.

OpenFOAM benchmark with settings of Input: drivaerFastback, Medium Mesh Size, Execution Time. acpi_cpufreq performance was the fastest.

The AMD P-State EPP power consumption was just a Watt or two lower than the ACPI CPUFreq results on this EPYC 7773X 2P server.

OpenFOAM benchmark with settings of Input: drivaerFastback, Medium Mesh Size, Execution Time. acpi_cpufreq performance was the fastest.

The ACPI CPUFreq and AMD P-State EPP modes were leading to slightly lower peak CPU clock frequencies than amd_pstate while still delivering that great performance.

Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi_cpufreq ondemand was the fastest.

With the Kvazaar open-source H.265 encoder there was more variation between the different CPU frequency scaling drivers/governors as we tend to see with the video encoders at large.

Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi_cpufreq ondemand was the fastest.
Kvazaar benchmark with settings of Video Input: Bosphorus 4K, Video Preset: Medium. acpi_cpufreq ondemand was the fastest.

Delivering the best performance-per-Watt now was the AMD P-State EPP driver with the powersave governor and balanced performance EPP preference.

SVT-AV1 benchmark with settings of Encoder Mode: Preset 13, Input: Bosphorus 4K. amd_pstate_epp performance balance_performance was the fastest.
SVT-AV1 benchmark with settings of Encoder Mode: Preset 13, Input: Bosphorus 4K. amd_pstate_epp performance balance_performance was the fastest.
SVT-AV1 benchmark with settings of Encoder Mode: Preset 13, Input: Bosphorus 4K. amd_pstate_epp performance balance_performance was the fastest.

With the SVT-AV1 open-source AV1 encoder from Intel there was also a tiny lead with AMD P-State EPP, this time when using the performance governor and balanced performance hint.


Related Articles