AMD Ryzen 9 7950X P-State/CPUFreq Frequency Scaling Performance On Linux

For not all workloads do the CPU frequency scaling driver/governor make a measurable difference on performance depending upon the CPU. For seeing acpi-cpufreq and amd-pstate tested across dozens of workloads, see this result page.

Across the 100+ tests conducted, here is a look at the peak CPU frequency across the tested combinations. Using amd-pstate performance led to the highest peak frequencies being achieved on the Ryzen 9 7950X and even a noticeable improvement over the acpi-cpufreq performance configuration. Running and-pstate schedutil led to a similar median peak frequency to acpi-cpufreq ondemand/schedutil but while also exhibiting a much lower peak frequency.

Here is a look at the CPU package power consumption expressed via RAPL across all of the tests carried out. Using amd-pstate schedutil led to a slightly lower average than acpi-cpufreq that was in surprisingly great shape.

The Ryzen 9 7950X did see some small differences to the core temperatures based on the drivers/governors. The amd-pstate schedutil thermal results were similar to acpi-cpufreq performance.

Running amd-pstate schedutil as is the default on most modern Linux distributions for Zen 2 and newer AMD CPUs tended to yield the slowest performance while to no surprise amd-pstate performance meant the best performance. That's where things stand with the AMD Ryzen 9 7950X across the different Linux driver/governor combinations. Frankly I'd love to see more Linux distributions go the route of Intel's Clear Linux where the performance governor is used by default on desktops/servers and then sticking to schedutil for laptops/mobile systems by default. The amd-pstate schedutil default isn't too compelling from my tests on the Ryzen 9 7950X and other AMD Ryzen processors, but we'll see how things change once the latest AMD P-State EPP driver code is ready for the mainline kernel next year.