14.04.2 & intel_pstate.

[faucon50]

Hi,

I came fortunatly accros this topic because I was surching the reason why my i7 was always at max cpu freq. My ondemand was wrong to, after the change to powersave everything seems to be right. Just to be sure, this part can be removed, isn't it ?

*ondemand*)
GOVERNOR="ondemand"
break
;;

This is the result on my setup buntu 14.04.2 :

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave

 cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq
2233105
1916113
1922363
800000
1599804
1265332
2500390
1078906

Thank you.

[Doug S]

I came fortunatly accros this topic because I was surching the reason why my i7 was always at max cpu freq. My ondemand was wrong to, after the change to powersave everything seems to be right. Just to be sure, this part can be removed, isn't it ?

*ondemand*)
GOVERNOR="ondemand"
break
;;

No, you can, and should, leave those lines included. When there is no "ondemand" then the script will fall through to the next case, "powersave" and do that part. By leaving the "ondemand"case intact you cover the situation of switching back to the acpi=cpufreq driver, or some future driver with something called "ondemand" or whatever.

[mc4man]

No, you can, and should, leave those lines included. When there is no "ondemand" then the script will fall through to the next case, "powersave" and do that part. By leaving the "ondemand"case intact you cover the situation of switching back to the acpi=cpufreq driver, or some future driver with something called "ondemand" or whatever.

I just put an example of proper ondemand script in to 1st. post
(- while it's easy to patch in place rather than explain how seems simpler to just give the whole script in a code box.

This issue & half a dozen others has me wondering why Ubuntu features the 14.04.2 HWE image on their page. Seems to be a 'not thought out at all' decision as that image is semi-broken in a number of areas. They should be using 14.04.2 'non HWE' with a link to HWE images for the few that need for a new install. Otherwise the decision to use a HWE packages should be left to users who can do so manually via upgrade/install from current 14.04.1/.2/.3/.4.

The use of unique named mesa packages should not be on the featured 14.04 image

[faucon50]

Ok, thank you for thoes precisions. I'm currently trying Bodhi linux 3 which is ubuntu 12.04.2 based and will inform them about that.

[faucon50]

Hi,

A "very" strange thing is that after the installation of the nvidia proprietary driver (311), I have no more the intel_pstate kernel driver but only the cpufreq and ondemand ! Is that normal ? My /etc/init.d/ondemand hasn't changed.

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
ondemand
ondemand
ondemand
ondemand
ondemand
ondemand
ondemand
ondemand

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq
800000
800000
800000
800000
800000
800000
800000
800000

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_driver
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq
acpi-cpufreq

Is it because of the nvidia driver and is there a away to revert to pstate ?

Thank you.

[Doug S]

Is it because of the nvidia driver

I don't know, I am a server guy and never had to be involved in what, on these forums, appears to be a never ending nvidia saga. However if it is the only change you made...

and is there a away to revert to pstate ?

I wonder if you can force it.

First, make a copy of your /etc/default/grub file, just in case:

sudo cp /etc/default/grub /etc/default/grub.original

then edit the file as sudo, for example (but you use your favorite editor):

sudo nano /etc/default/grub

and add this to the GRUB_CMDLINE_LINUX_DEFAULT line:

GRUB_CMDLINE_LINUX_DEFAULT="intel_pstate=enable"

If there are already other parameters on that line, then add the new one. Example:

GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 intel_pstate=enable crashkernel=384M-:128M"

After you save the file, then update grub

sudo update-grub

Re-boot the computer.

[faucon50]

Yes, "intel_pstate=enable" option did it :

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_driver
intel_pstate
intel_pstate
intel_pstate
intel_pstate
intel_pstate
intel_pstate
intel_pstate
intel_pstate

 cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave

 cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq
806250
806250
806250
806250
806250
806250
806250
806250

But it seems to be a strange behaviour, a mix between ondemand and pstate because in idle all cores are at the same freq which was not the case before... !?

[Doug S]

But it seems to be a strange behaviour, a mix between ondemand and pstate because in idle all cores are at the same freq which was not the case before... !?

Your system wasn't really "idle" before, and now it appears that it really was "idle" when you executed the command.

Is your system a desktop system with a GUI? If yes, then they have all kinds of stuff running and there really is no such thing as "idle". On a server system with no GUI, there is a lot less stuff running and while "idle" still doesn't really mean "idle", it tends to be an awful lot closer.

[faucon50]

Yep still strange, because as you can see few posts ago, the driver was p_state and the governor powersave and not "really" idle as now in the same environment... Maybe the reason is the interaction intel CPU/nouveau driver vs intel CPU/nvidia propritery driver ? Now it's almost constantly at 806 MHz which looks like "real" idle.

I'm on a laptop (alienware 14) with Bodhi linux (buntu 14.04.2 based & enlightenment GUI). It's clear that it as nothing to do with a server. Normally I use to install buntu server and then Xfce4 over, which is not that over sized as a regular (x,u,l)buntu and 350 eyes kandy useless things.

What about themald ? Few months ago it was a bit buggy on buntu. Did someone of you tried ?

[Doug S]

What about themald ? Few months ago it was a bit buggy on buntu. Did someone of you tried ?

I'm still hearing about situations where thermald causes issues, but don't have any real proof. I think I will just have to install it and play with it. Even when all my CPUs are at full load, my test computer doesn't get hot enough for it to kick in, so I will have to make lower trip points to test. I am finding the man page for the config file a little useless because it doesn't define the units of things. Example:

<SamplingPeriod> 12 </SamplingPeriod>

what are the units of "12"? For goodness sakes.