Intel Skylake Performing Better With CPUFreq Than P-State For Linux Gaming

[phoronix]

Phoronix: Intel Skylake Performing Better With CPUFreq Than P-State For Linux Gaming

While the latest Linux kernel has Intel P-State driver support for CPU frequency scaling being done by this Intel-specific driver rather than the common ACPI CPUFreq driver, for some common tests -- especially Linux gaming -- the latest-generation Intel CPUs perform noticeably better using the CPUFreq driver. Here are some CPUFreq vs. P-State drivers when testing the available scaling governors.

http://www.phoronix.com/vr.php?view=22261

[brauliobo]

there is something wrong with the benchmarks, there is no difference between powersave and performance, even for cpu intensive benchmarks. the cpufreq setting is not working

[andrei_me]

I would like to see the power consumption along with these tests, to see if there is any trade-off on performance per watt

[Michael]

I would like to see the power consumption along with these tests, to see if there is any trade-off on performance per watt

My lone WattsUp meter was busy on another system.

[Michael]

I would like to see the power consumption along with these tests, to see if there is any trade-off on performance per watt

There's nothing wrong with the benchmarks as time and time again they've proven to be reproducible, must be something going on with the CPU, microcode, or the motherboard UEFI.

[s_j_newbury]

Recent Intel CPUs/chipsets have a lot of power management control between different units. Does that work at all with the cpufreq driver? Does perf-bias make a difference? Even on my IvyBridge system perf-bias makes a huge difference to performance/power consumption, it's my understanding newer chips have even greater potentials in this regard.

[Guest]

There's nothing wrong with the benchmarks as time and time again they've proven to be reproducible, must be something going on with the CPU, microcode, or the motherboard UEFI.

https://www.kernel.org/doc/Documenta.../governors.txt clearly states that "The CPUfreq governor "powersave" sets the CPU statically to the lowest frequency within the borders of scaling_min_freq and scaling_max_freq."

Find the problem and fix it!

Good luck.

[CrystalGamma]

I would like to see the power consumption along with these tests, to see if there is any trade-off on performance per watt

Me too. I'd also like to see the idle power for each of the setups.

[philips]

there is something wrong with the benchmarks, there is no difference between powersave and performance, even for cpu intensive benchmarks. the cpufreq setting is not working

Indeed weird, but the Phoronix suite is working alright, here's what I've got with two tests:

[Michael]

https://www.kernel.org/doc/Documenta.../governors.txt clearly states that "The CPUfreq governor "powersave" sets the CPU statically to the lowest frequency within the borders of scaling_min_freq and scaling_max_freq."

Find the problem and fix it!

Good luck.

Yep, glad to with enough subscripbers/support.