Ryzen Mobile Power/Performance With Linux 6.3's New AMD P-State EPP Driver

As with the AMD EPYC testing, the AMD Ryzen Mobile "Rembrandt" laptop didn't see all that big of a difference for some workloads... With the strong exception of acpi-cpufreq and amd_pstate with the powersave governor. The powersave governor in those combinations led to multiple times slower performance than the other combinations tested. Meanwhile when it came to measuring the time to compile the Linux kernel, any of the other driver/governor combinations were basically leading to similar results.

The AMD P-State / CPUFreq powersave reuslts are so slow as with amd_pstate powersave, even when having multi-threaded jobs the Ryzen 7 PRO 6850U wasn't running above 400MHz on any of its 8 cores... With ACPI CPUFreq powersave, the 6850U wasn't hitting above 1600MHz. With the other combinations tested, they were running at 3.1~3.3GHz for their average peak and topping out at 4.7GHz when boosting all the way.

Running with the amd_pstate powersave and acpi-cpufreq powersave modes does lead to much less power consumption, but build times are painfully slow. At least with amd_pstate_epp, the powersave mode isn't so painfully slow and led to slightly better power efficiency than the other tested combinations.

The amd_pstate_epp powersave power mode also led to slightly better CPU core temperatures than the rest -- not counting the poor amd_pstate/acpi_cpufreq powersave modes.

Throughout all the code compilation tests and other intensive workloads, the powersave mode with ACPI CPUFreq and passive AMD P-State was very slow. At least though the new amd_pstate_epp driver with its powersave mode was being competitive performance-wise and slightly ahead the other combinations.