Feral's GameMode 1.1 Released For Optimizing Linux Gaming Performance

[geearf]

As you say 2w is not useless on a major scale so even if you have a single computer, there exists probably millions in the country where you live so the energy savings for a society as a whole can also be quite significant even if the save is only 2w per system.

Sorry about the "might not equal that of others", hadn't followed your posts back to your first one.

The diff was only in mprime, which is using the CPU to the max (well sort of on option 3...) at that point with gamemode you'd be running performance anyway, so you wouldn't save anything... Actually, from what you're saying, by not using gamemode and sticking to powersave, you'd be better off.

[F.Ultra]

The diff was only in mprime, which is using the CPU to the max (well sort of on option 3...) at that point with gamemode you'd be running performance anyway, so you wouldn't save anything... Actually, from what you're saying, by not using gamemode and sticking to powersave, you'd be better off.

To be completely honest I do not know, I have not made any power measurements between Ondemand and Performance for decades now so my "knowledge" can very much be outdated. One might think that distributions defaults to Ondemand for a reason though.

edit: However running the cpu at 100% should show no difference between the two governors, the major differences should be with applications that create a varying stress on the CPU where the Ondemand governor will keep the frequency at a lower rate (by smoothing out the spikes so to speak) than the Performance governor.

[geearf]

To be completely honest I do not know, I have not made any power measurements between Ondemand and Performance for decades now so my "knowledge" can very much be outdated. One might think that distributions defaults to Ondemand for a reason though.

edit: However running the cpu at 100% should show no difference between the two governors, the major differences should be with applications that create a varying stress on the CPU where the Ondemand governor will keep the frequency at a lower rate (by smoothing out the spikes so to speak) than the Performance governor.

Oh as said before I'd assume cpufreq governors to behave quite differently than pstate ones.
schedutil may be more sane powerwise than ondemand or performance, but I have not tested any of these so this is just a guess out of not much.
I could if you wanted too, though I wouldn'd do as many tests as last time, as it takes a while...

As for no difference at 100%, yes that's also what I'd expect with pstate, but not with cpufreq powersave vs performance. Are the 2w any meaningful since it's the full draw at the PSU and not just the CPU, and it's not even 2% difference? I am not sure, but it's interesting that I had the same difference on my 3 perf_bias tests. I think the previous time I did not notice any difference on max but on min by 1-3w, not sure why it changed... or if it's meaningful at all as it's so tiny.


As for the default, well it's only the good one till proven it is not
As for ondemand, it definitely is not the default for newer Intel CPUs.. and maybe schedutil will become the cpufreq default once matured enough.
A while back I had some issues with schedutil and interacted with a dev about this (sorry forgot whom), at the end I asked what I should use and he said either schedutil or pstate's powersave, that they would be pretty equivalent in results, at least for my use case.

I think at this point, we got quite far from the original question I was trying to answer