Intel P-State vs. CPUFreq Benchmarks On The i7-5960X

[phoronix]

Phoronix: Intel P-State vs. CPUFreq Benchmarks On The i7-5960X

Continuing in our recent CPUFreq vs. P-State scaling driver benchmarks, here's some tests from the eight-core Core i7 5960X Haswell-E system as we test the two CPU frequency scaling drivers and their different governors.

http://www.phoronix.com/vr.php?view=20920

[rudl]

Benchmarks won't show the real improvements that clearly, I suggest everyday usage scenarios with cpu loads under 50% per core with clockspeed monitoring and subjective performance/lag.
like a flash-video. browsing in gerneral, Or games with low settings and a framecap/vsync.

[Apopas]

I think I'm gonna switch from conservative to powersave!

[artivision]

I think I'm gonna switch from conservative to powersave!

x86 mobile is a joke, the hole market is a lie. I always have 70% more performance on Wine and all emulators when i am on performance. But my laptop will eventually fail.

[dungeon]

On my Kabini i disable cpufreq governor for good (i mean my kernel does not even have it compiled) and there is no any difference in power consumpation, temperature and fan speed . What a joke is that

Anyway seems to me newer AMD APUs also need pstate like governer, but no one implemented it - Intel, can we reuse your one? . I know some? AMD kernel CPU people has been laid off few years ago, but things now starts to really suffer in that area... not good

[kokoko3k]

Timed Linux Kernel Compilation v3.1

P-State Powersave: 51.11 secs
P-State Performance: 57.6 secs

Powersave was faster then Performance, is this a bug or what?

Also...
Less than a minute to compile the kernel!?!?!?

[profoundWHALE]

Also...
Less than a minute to compile the kernel!?!?!?

$1000 CPU delivers $1000 CPU performance. Don't know why you're surprised.

[liam]

Timed Linux Kernel Compilation v3.1

P-State Powersave: 51.11 secs
P-State Performance: 57.6 secs

Powersave was faster then Performance, is this a bug or what?

Also...
Less than a minute to compile the kernel!?!?!?

Yeah, that one caught my eye as well. This MIGHT be due to the difference between how the two work rather than a bug, exactly. Kernel compilation is a good benchmark b/c it is a really heavy load on the entire system. Given this is being run with an ssd(the reason for the fast compilation is imagine) this hit might be the result of the performance governor ramping up and down so much (down when doing io which, for ssd, is on the high usecs), while powersave stays on constantly.
Obvs not sure but that was what came to mind.

Edit: Just noticed that acpi perf and conservative (I'm assuming powersave should have similar heuristics to conservative) acted more reasonably do I'm changing my answer to bug.

[Xelix]

Michael,

Interesting findings. Have you considered investigating those results? Instead of just dumping the raw results from your (great) benchmark suite, why not spend the N days necessary to understand why the performance results are what they are?

[samdraz]

Can you redo the test with thermald https://01.org/linux-thermal-daemon on computer, plain p-state driver installation wont help that much,