[Solved]Kernel 3.12, Haswell, and Intel Power Management

BahbGTR · 2013-11-19 13:17:20

I guess I have rotten luck when it comes processor power management.

Pertinent Specs:
Xeon 1245v3
Asrock z87 Extreme6 (booting in UEFI mode)
Nvidia GTX 760 (running latest stable nvidia drivers)
1 SSD
3 HDDs
1 BD-R

Today I noticed my system making more noise than usual (fan noise) so I checked out i7z. Here is what I get with the 3.12.0-1 repo kernel

Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 3399 MHz
  CPU Multiplier 34x || Bus clock frequency (BCLK) 99.97 MHz

Socket [0] - [physical cores=4, logical cores=8, max online cores ever=4]
  TURBO ENABLED on 4 Cores, Hyper Threading ON
  Max Frequency without considering Turbo 3498.97 MHz (99.97 x [35])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4 Cores is  38x/38x/37x/36x
  Real Current Frequency 3474.61 MHz [99.97 x 34.76] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %
        Core 1 [0]:       3419.96 (34.21x)         1       5.6      0      0
        Core 2 [1]:       3416.03 (34.17x)         1     3.08      0       0
        Core 3 [2]:       3474.61 (34.76x)         1     28.7      0       0
        Core 4 [3]:       3471.60 (34.73x)         1     1.09      0       0

Notice that the cores are running full tilt. This is with 1-2% cpu utilization, about 10 minutes after boot. Same behavior after being on for hours too, even with low CPU usage. I tried rebooting without the nividia modules and virtualbox host modules, and I get the same behavior.

So for fun, I tried the LTS kernel (3.10.19-1-lts). Here is what I get with i7z

Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 3398 MHz
  CPU Multiplier 34x || Bus clock frequency (BCLK) 99.94 MHz

Socket [0] - [physical cores=4, logical cores=8, max online cores ever=4]
  TURBO ENABLED on 4 Cores, Hyper Threading ON
  Max Frequency without considering Turbo 3497.94 MHz (99.94 x [35])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4 Cores is  38x/38x/37x/36x
  Real Current Frequency 810.01 MHz [99.94 x 8.10] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %
        Core 1 [0]:       799.39 (8.00x)           1    8.07       0       0
        Core 2 [1]:       810.01 (8.10x)           1    3.11       0       0
        Core 3 [2]:       799.52 (8.00x)           1    0.422      0       0
        Core 4 [3]:       799.88 (8.00x)           1    8.97       0       0

Power management is working as advertised with the LTS kernel.

Anyone else seeing this behavior too? This worked just fine with kernel 3.11.X. I tried ruling out other bits of software by booting the system without X, GPU drivers, etc to rule out other possibilities.

Edit:

I've found that the LTS kernel uses a different governor than the intel pstate driver with my specific chip which explains the difference in operating frequencies between the two. So disabling pstate would get the same results, but that's not really something I'm interested in doing.

Last edited by BahbGTR (2013-11-20 15:52:00)

Roberth · 2013-11-19 14:13:41

Use cpupower to try out different governors, afaik, pstate's governors works differently.

But out of curiosity, why are you so concerned about this on an desktop computer?

BahbGTR · 2013-11-19 14:33:20

Just because it's a desktop doesn't mean I'm not concerned about power consumption. I'm kind of cheap like that

If a processor has a feature, I want it to work. I've had issues with pstate before, but it seemed to work OK for a bit.

WonderWoofy · 2013-11-19 15:02:31

I would be more worried about the fact that it is not reaching all the sleep states rather than the actual frequency of the processor. Being in a lower frequency will indeed lower the power consumption, but with these new processors, there is no tool that can tell you precisely what it is doing all the time, since the frequency variability works in the realm of microseconds. So instead, you should hopefully see that your processor is going into the c3 and c6 states. With Haswell though, I know that there are new sleep features that have been implemented which I don't think have been implemented yet in i7z (or powertop).

It is not often that there are big issues with the intel_pstate driver, but they do occur, and since the release of the 4th gen processors, it is no surprise that many of those sleep issues have been with Haswell (they are new after all). Maybe in the mean time, you might want to disable the intel_pstate driver and use the old acpi-cpufreq and ondemand. You can do this by adding 'intel_pstate=disable' to the kernel command line (it might be intel-pstate=disable I can't remember, nor do I need to use this).

brain0 · 2013-11-19 15:06:52

BahbGTR wrote:

Here is what I get with the 3.12.0-1 repo kernel

Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 3399 MHz
  CPU Multiplier 34x || Bus clock frequency (BCLK) 99.97 MHz

Socket [0] - [physical cores=4, logical cores=8, max online cores ever=4]
  TURBO ENABLED on 4 Cores, Hyper Threading ON
  Max Frequency without considering Turbo 3498.97 MHz (99.97 x [35])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4 Cores is  38x/38x/37x/36x
  Real Current Frequency 3474.61 MHz [99.97 x 34.76] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %
        Core 1 [0]:       3419.96 (34.21x)         1       5.6      0      0
        Core 2 [1]:       3416.03 (34.17x)         1     3.08      0       0
        Core 3 [2]:       3474.61 (34.76x)         1     28.7      0       0
        Core 4 [3]:       3471.60 (34.73x)         1     1.09      0       0

Notice that the cores are running full tilt. This is with 1-2% cpu utilization, about 10 minutes after boot.

That is completely normal behaviour. This post is by an intel engineer who explains this. So basically, there is no need to clock down your CPU below the highest non-turbo frequency.

I have no idea how that relates to your fans, just confirming that the CPU side of things is completely normal.

As WonderWoofy points out, you should be worried about the C states, not the P states - however, I also have no idea if the reporting of the P states is correct.

Last edited by brain0 (2013-11-19 15:09:22)

BahbGTR · 2013-11-19 15:12:56

So it's a non-issue then. I can handle that. I'd rather have access to the turbo frequencies than use the other governors. Guess I'll carry on then.

Thanks all!

blackout23 · 2013-11-19 22:34:41

Don't forget to mark your thread as [solved]

Gusar · 2013-11-19 22:54:51

That i7z output is bogus, because you're using a version that doesn't have Haswell support. Use i7z-git from AUR, that one will show all C states correctly. Powertop should also work, and then there's turbostat.

Last edited by Gusar (2013-11-19 23:00:16)

BahbGTR · 2013-11-20 15:51:20

I actually am using the i7z-git package.

pacaur -s i7z-git
aur/i7z-git f07785f-1 (3) [installed]
    A better i7 (and now i3, i5) reporting tool for Linux. Git version.

I didn't realize that the output didn't fit in a standard 80x25 console so I never knew about the other columns. But here is the full output that does confirm things are working as intended with the C7 state

Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 3398 MHz
  CPU Multiplier 34x || Bus clock frequency (BCLK) 99.94 MHz

Socket [0] - [physical cores=4, logical cores=8, max online cores ever=4]
  TURBO ENABLED on 4 Cores, Hyper Threading ON
  Max Frequency without considering Turbo 3497.94 MHz (99.94 x [35])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4 Cores is  38x/38x/37x/36x
  Real Current Frequency 3482.08 MHz [99.94 x 34.84] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %   C7 %  Temp      VCore
        Core 1 [0]:       3419.34 (34.21x)         1    6.14       1       1    91.5    25      1.0565
        Core 2 [1]:       3406.66 (34.09x)         1    0.463      0       0    99.4    23      1.0565
        Core 3 [2]:       3482.08 (34.84x)         1    3.02       0       0    96.9    22      1.0564
        Core 4 [3]:       3467.04 (34.69x)         1    3.05       0       0    96.9    25      1.0593

Last edited by BahbGTR (2013-11-20 16:00:57)

WonderWoofy · 2013-11-20 16:11:05

Yeah that looks super normal to me!

jancici · 2013-11-22 13:58:31

HI

I have i7-4750HQ in my laptop.

I find this helpfull. So I did instal i7z from AUR but still not see C7 state.

pacman -Qs i7z
local/i7z-git 5023138-1

I dont know how to copy paste output from i7z here. But it is almost same only I have two more lines at top:

Cpu speed from cpuinfo 1995.00Mhz
cpuinfo might be wrong if cpufreq is enabled. ....

well, I did install cpupower but dont know if it has sense.
maybe I have something wrong, but dont know ...

when running powertop I can see that all 4 cores are at C7s-HSW state 97%

thanks forhelp toget i7z showing that C7 state. maybe it will help me save battery

Arch Linux

#1 2013-11-19 13:17:20

[Solved]Kernel 3.12, Haswell, and Intel Power Management

#2 2013-11-19 14:13:41

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#3 2013-11-19 14:33:20

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#4 2013-11-19 15:02:31

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#5 2013-11-19 15:06:52

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#6 2013-11-19 15:12:56

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#7 2013-11-19 22:34:41

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#8 2013-11-19 22:54:51

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#9 2013-11-20 15:51:20

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#10 2013-11-20 16:11:05

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

#11 2013-11-22 13:58:31

Re: [Solved]Kernel 3.12, Haswell, and Intel Power Management

Board footer