AMD Ryzen "Renoir" CPU Frequency Scaling Governor Performance

The benchmarks shown aren't even the full data set, but by now you get the point. Those wanting to see all 129 test results along with all of the sensor outputs on a per-test basis can visit this OpenBenchmarking.org result file. Many different workloads tested so hopefully covering all of the different areas that may be important to you and your use-cases.

But in providing a concise overview, here are the results on a macro level over the span of all 129 tests carried out:

On average during the 129 tests, the Ryzen 5 4500U had an average power consumption reported by the new amd_energy driver of 13~14 Watts and a peak of 30 Watts. The only CPUFreq governor configuration with noticeably different results was powersave where the average load CPU power consumption was 5.5 Watts with a peak of 15 Watts.

But even with the powersave governor at nearly a third of the power consumption rate as the other governors, it isn't too practical. When taking the geometric mean of all 129 tests against the average power consumption rate, the powersave efficiency still falls well short while the "performance" governor has a narrow lead over ondemand, schedutil, and conservative.

The powersave governor was successful in keeping the Lenovo laptop running cooler with a core temperature of 42 degrees compared to an average temperature of 76~78 degrees with the other governors. Similarly, the peak CPU temperature during testing was 56 degrees with powersave but 103 degrees reported for the others.

Using powersave also kept the GPU temperature lower with an average of 40 degrees compared to 62~64 degrees.

The overall system temperature also reported much lower results with powersave while not much of a difference between the other governors exposed by the CPUFreq driver.

If caring about just maximizing the raw performance out of your modern AMD Ryzen "Renoir" laptop, using the performance governor led to around 7% higher performance compared to the common default of "ondemand". Meanwhile schedutil came in second place with a faint lead over ondemand while performance was still 5% faster overall than that modern governor employing the kernel's scheduler utilization data.

The main exception to "the performance governor delivering the best performance" was with the graphics tests. As seen also with Intel CPUs and their integrated graphics, using the more conservative governors depending upon the game or visual application can lead to higher frame-rates due to the CPU/GPU sharing the same thermal/power envelope. That was the case here with the Ryzen 5 4500U but it really comes down to the graphics application/game for how much of a difference is made.

If you enjoyed these benchmarks consider showing your support by joining Phoronix Premium and stop by OpenBenchmarking.org for all of the data in full.