The intel_pstate scaling driver module seems to be missing completely. Has it been removed for having problems or just missed in the latest config?



Also, there's no pstate governor folder present at all.




< Edited by gvoima :: Jan 6, 14, 7:40 >
Back to top

Yes, it's been removed. From the debian changelog:



I don't think this governor is ready. Even Valve did testing and removed it from the kernel with good reasons:
Source: www.phoronix.com/scan.php?page=news_item&px=MTU0Mzc

Besides that, there is a thread in this forum talking about its problems. The SteamOS citation was the last straw - I'll put it back in when it's proven to improve performance _and_ reduce power consumption.
Back to top

Seems that my temperatures haven't changed a bit between governors, and I don't see any performance differences between ondemand and pstate.

Wine segfaults a lot, but that's a whole 'nother story..
Back to top

I assume we can still turn it back on if we build manually?

I have not had problems with Intel P-State on my 2600K Sandybridge.
It is more responsive than ondemand for me at least since it clocks up and down much faster than ondemand.
I do think it works better with turbo disabled and a regular overclock though.

I have not had time to update from 3.10 since a Nvidia driver for 3.11 or 3.12 came out so there could have been problems introduced since then I have not had a chance to observe yet.

< Edited by arclance :: Dec 23, 13, 16:48 >
Back to top

gvoima, that's a good thing that you can't tell any difference - that just means that you were unaffected. It appears that there are problems with Intel P-State on just a very specific hardware configuration or CPU type.

I'm going to wait it out and watch the news on Intel P-State very closely.

arclance, yes you can re-enable it if you turn it back on in the configuration.
Back to top

I had time to build 3.12.6-1 today and turned on Intel P-State.
It works fine for me, a little better than 3.10.12 actually.


For your reference these are the details of my system.

I have turbo disabled on the CPU and the max clock speed set to 4.1 GHz (0.7 GHz overclock) with a aftermarket cpu cooler.
This is the ram I am using.
I have the power level fixed at max on the GPU because the power levels change too slowly and cause stuttering if it is left on.
Back to top

arclance, what is better? Can you do a kernel build benchmark? (make allnoconfig && time make -j10) and check your CPU temperatures when idle?

If everything works as intended, you should get higher throughput and identical temperatures. However, what more people are experiencing is identical or faster throughput and higher temperatures.

Even though coretemp's data is all over the place, I can still see that my idle temperature with a few things running is much lower than with intel p-state:



It was impossible to get these temperatures without turning everything off. That ruins the point of a computer though - you use it to do work or play games on.

This is not a very scientific example, but if you can prove to me that the p-state cpufreq driver is not responsible for the higher temperatures, then I'll re-evaluate intel p-state.
Back to top

I could do that this weekend most likely.
I don't have time today or tomorrow to do things that require reboots (turning off intel p-state) until then.

I don't know how accurate it would be though since my fans are PWM and the motherboard just speeds them up (all 9 of them) as things gets hotter.
I don't think a kernel compile will take long enough to overcome the fans ability to compensate for a minor change in thermal output though.

It is possible that because of that my fans are running faster so there is no noticeable change in measured temp for me even if the CPU is actually running hotter.

We can compare CPU temp and fan speed to check that though.

This is what it looks like right now just typing in iceweasel.

Core 3 always reads higher than the others (the difference is less under a heavier load), I don't know why.

This computer sits at ~32-36 C (~35-40 C during the hottest part of summer) under normal use (web browsing, playing HD videos, and using a text editor at the same time) using intel p-state.
After 45min-1hour under full load it gets to ~45-55 C and tops out at ~55-65 C after more than 4 hours depending on the time of year.
My case is setup for light air cooled overclocking though so these temperatures might not be typical.

I am also not using the GPU built into the processor so I will see lower temps than people who do use it.

I am using a much lighter system than most people do (fluxbox, using startx instead of a graphical login manager so I could strip out as many unneeded things as possible) so that may be a factor how low my temps are as well.
Back to top

Hi!

Since acpu_cpufreq supports "boost", it's better than intel_pstate. My computer is running cooler than with pstate and obviously battery time is better.
As an example, with pstate almost always cpu is running in turbo mode, for me ~ 2.8Ghz (nominal is up to 2.1GHz), but with cpufreq it's about 1.4GHz. But with either of the drivers, states > C0 are used, the difference is MHz only.


Back to top