Intel Xeon Platinum 8380 P-State CPU Frequency Scaling Comparison

Written by Michael Larabel in Software on 7 June 2021. Page 6 of 6. 6 Comments

In total there were 100 benchmarks run across the four governors on this Intel Xeon Platinum 8380 dual socket server running Ubuntu 21.04 Linux.

Across that wide range of different workloads, overall switching to the performance governor from the default powersave governor meant an increase of ~27% based on the geometric mean. Interestingly even using the Schedutil governor meant a 7% improvement for the geometric mean across all of the benchmarks tested.

For the ~15 hours it took each time to run all 100 benchmarks, here is a look at the peak CPU frequency observed each second from any of the CPU cores. The performance governor meant an average peak frequency of 3.1GHz while the schedutil governor averaged out to 3.0GHz and the powersave default was at around 2.8GHz.

When looking at the combined CPU package power consumption over the entire run for each configuration, the powersave run yielded a 371 Watt average for the two Xeon Platinum 8380 processors while the Schedutil governor came out slightly ahead at 368 Watts. When using the performance governor was about a 12% increase in power consumption with a 413~417 Watt average. But considering the 27% improvement in the geometric mean, not bad on average.

On the thermal front, the performance governor led to the CPU cores operating just a few degrees higher on average and for their peak frequencies while understandably the minimum observed core temperature was higher with the performance governor keeping to the higher performance states.

Those wanting to dig through all 100 benchmarks in full plus the various per-test power/efficiency metrics can do so via this OpenBenchmarking.org result file.

That's the short story on how the P-State driver and governor performance is looking for the dual Xeon Platinum 8380 Ice Lake. It's still somewhat surprising that for servers at least there aren't more Linux distributions defaulting to the performance governor given the increasing impact with newer generations of CPUs. When the system isn't fully saturated for prolonged periods of time or just going through bursts of heavy load, as we have shown before and again with these results the powersave governor can be a poor choice in such situations. Or at least not a more aggressive powersave governor for today's hardware with even Schedutil providing a better showing than powersave. With CPUFreq on AMD systems with modern kernels, Schedutil is the default there so perhaps in the future we'll see it also be considered on the Intel side.

For those wondering about P-State with current Intel desktops, back in April I ran some P-State tests with Rocket Lake / Core i9 11900K.

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.


Related Articles
About The Author
Michael Larabel

Michael Larabel is the principal author of Phoronix.com and founded the site in 2004 with a focus on enriching the Linux hardware experience. Michael has written more than 20,000 articles covering the state of Linux hardware support, Linux performance, graphics drivers, and other topics. Michael is also the lead developer of the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software. He can be followed via Twitter, LinkedIn, or contacted via MichaelLarabel.com.