[jphalimi]

Absolutely. I don’t know what this tool is about, but the README statement is wrong; I can’t think of any distribution dumb enough to use performance governor by default. Most Intel-based platforms use Intel-specific governors.

I developed an energy-aware DVFS controller for HPC clusters a few years back, and there was no such thing as performance by default even on compute nodes.

[bratch]

Intel's own distribution defaults to the peformance governor [1] in order to race to idle. This has been Intel's recommendation for some years.

[1] https://docs.01.org/clearlinux/latest/guides/maintenance/cpu...

[kylek]

Interesting note that I wasn't aware of-

>> The intel_pstate driver only supports performance and powersave governors.

What happens when ondemand is used with the intel driver? (is it even selectable?)

[Aengeuad]

  # echo performance | tee /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
  performance
  # echo ondemand | tee /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
  ondemand
  tee: /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor: Invalid argument
  # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
  performance

The Arch Wiki suggests that both powersave and performance have dynamic scaling, and running 'watch -p -n 0.25 grep \"cpu MHz\" /proc/cpuinfo' indicates that both of them do, with powersave mostly sitting at 1.2GHz with occasional increases on a handful of cores while powersave fluctuates from 2~ to 4.2GHz across all cores with minimal load.

The linked docs from Clear Linux say that power draw isn't entirely dependant upon cpu frequency when there's no load so there's no issue with keeping it on performance, and I'm not about to doubt Intel here, but I'd be surprised if powersave didn't save power if only by clamping down how many resources programs can use. Anecdotally I've noticed that the powersave governor doesn't really work too well when doing things like running virtual machines and will keep the frequency very low, as if the scaler is blind to the resources the VM is using, while having it on the performance governor will pin all my cores to a far more appropriate 4.2GHz.

[eqvinox]

It's not selectable:

  /sys/bus/cpu/devices/cpu0/cpufreq # echo -n 'ondemand' > scaling_governor 
  echo: write error: invalid argument

Without touching anything, it's using powersave on my box. It seems to work well enough, both on performance as well as battery life, but that's just my personal experience.

The question that this tool completely fails to consider is that clocking the CPU lower isn't automatically the most battery efficient thing to do, cf. race to idle mentioned above. To me, this sounds a lot like someone on the trough of the Dunning-Kruger curve looked at a problem and wrenched at it without understanding enough of it.