AMD P-State vs. ACPI CPUFreq Testing With Ryzen Laptops On Linux 5.17

Over the entire duration of different benchmarks carried out, here is how the peak CPU frequency looked over the span of benchmarks ran in all the tested hardware/software combinations.

And the CPU power consumption over the entire span of benchmarks. On the lower-power Ryzen 5 5500U there wasn't much of a difference to report. In the case of the AMD Ryzen 9 5900HX with ASUS laptop, the AMD P-State Schedutil driver did lead to around 3 Watts lower CPU power consumption on average than the standard ACPI CPUFreq Schedutil driver. When using the AMD P-State driver with performance governor it was 12 Watts lower on average than the ACPI CPUFreq Performance combination, but that seems to be an artifact of the amd_pstate driver sometimes making bad decisions given that the "performance" power consumption numbers were lower overall than even the "schedutil" run.

The lower power consumption when using AMD P-State on the high-end Ryzen 9 5900HX laptop also translated to slightly lower CPU core temperatures.

But when taking the geometric mean of all the raw performance benchmark results, the AMD P-State driver overall came just behind the ACPI CPUFreq driver on these two tested modern AMD laptops. So overall the AMD P-State driver didn't prove to be particularly compelling for its initial premiere in Linux 5.17. I'll continue testing AMD P-State on more laptop/desktop hardware and seeing how ongoing kernel improvements will impact it over the longer-term with its goal of delivering better energy efficiency for AMD Ryzen hardware on Linux. For right now though it would seem most users are still best off relying on the existing acpi_cpufreq driver.