Feral's GameMode 1.1 Released For Optimizing Linux Gaming Performance

[phoronix]

Phoronix: Feral's GameMode 1.1 Released For Optimizing Linux Gaming Performance

One month ago Linux game porter Feral Interactive introduced GameMode as a utility/service for dynamically optimizing the Linux system performance when running games. The initial focus on GameMode was on ensuring the CPU scaling governor was in its performance mode while today brought the GameMode v1.1 release...

http://www.phoronix.com/scan.php?page=news_item&px=Feral-GameMode-1.1-Released

[humbug]

Good stuff.

But I hope that the Linux kernel and the default configurations of popular distros head in a direction which makes this thing redundant soon.

There should be no need for this.

[Veto]

.deb or it didn’t happen

[aufkrawall]

But I hope that the Linux kernel and the default configurations of popular distros head in a direction which makes this thing redundant soon.

I suppose the shitty intel_pstate=powersave governor will remain the default for Intel CPUs for years (decades?).
Btw: schedutil got totally messed up here with 4.17, I suspect due to the reduced idle loop overhead.

That tool might be nice for newbies, but actually there is all you need to know inside the Arch Wiki. With modern CPUs inside of desktop systems it neither should be necessary to ever switch to something else than acpi-cpufreq=performance. My 6700k reaches ~25°C idle temperature with it (air cooling with ~320rpm)...

[R41N3R]

I suppose the shitty intel_pstate=powersave governor will remain the default for Intel CPUs for years (decades?).
Btw: schedutil got totally messed up here with 4.17, I suspect due to the reduced idle loop overhead.

That tool might be nice for newbies, but actually there is all you need to know inside the Arch Wiki. With modern CPUs inside of desktop systems it neither should be necessary to ever switch to something else than acpi-cpufreq=performance. My 6700k reaches ~25°C idle temperature with it (air cooling with ~320rpm)...

And what about power consumption? I think this should be noticeable as the CPU wouldn't reduce Mhz and use the sleep states.

[Venemo]

Why not just use the tuned daemon for this? It has been around for several years and has the same features and a lot more.

[aufkrawall]

And what about power consumption? I think this should be noticeable as the CPU wouldn't reduce Mhz and use the sleep states.

The secret is the CPU internally uses lower clocks than reported to the OS (when C-States are configured correctly in the bios), otherwise the power consumption would be higher and thus the low idle temperatures wouldn't be possible.
The reported CPU clock in OS is ~4GHz in idle, but power consumption of my whole system is just ~28W (RX 560 with idle clocks included). So definitely no need to ever use a CPU clock governor again for power saving purposes, at least with a stationary system.

[elvenbone]

The secret is the CPU internally uses lower clocks than reported to the OS (when C-States are configured correctly in the bios), otherwise the power consumption would be higher and thus the low idle temperatures wouldn't be possible.
The reported CPU clock in OS is ~4GHz in idle, but power consumption of my whole system is just ~28W (RX 560 with idle clocks included). So definitely no need to ever use a CPU clock governor again for power saving purposes, at least with a stationary system.

Which CPU intentionally reports false clock speeds to the OS?

[geearf]

And what about power consumption? I think this should be noticeable as the CPU wouldn't reduce Mhz and use the sleep states.

When game mode came out I did a simple test at the PSU with pstate and my i7.
At almost idle, the difference between performance and powersave was on average 1w, max 3w.
At full CPU load using mprime, it seemed similar between the 2 (and quite higher than above obviously).
Now my CPU does not throttle on performance while others do, so their difference may even be less than mine.

I'm guessing for an average user with one computer, at least with a similar configuration to mine, the whole debate of consumption with powerstate vs performance is irrelevant. For a farm with thousands upon thousands of CPUs and relatively little else in the computers, saving a watt here a watt there could be nice.

cpufreq might be a total different story, I didn't measure it, but I could if there's interest.

[aufkrawall]

Which CPU intentionally reports false clock speeds to the OS?

Likely every one which uses correctly functioning internal clock adjustment without slowing OS governor, so everything since Skylake and Ryzen.