View previous topic :: View next topic |
Author |
Message |
ayvango Tux's lil' helper
Joined: 08 Feb 2012 Posts: 118
|
Posted: Wed Sep 11, 2013 3:19 am Post subject: how should new cpupower be used |
|
|
I've read from several topics that cpufrequtils should be discarded for sake of cpupower utility. I've installed it and got some troubles.
There is no config file in /etc that describe cpupowerd behaviour like it was done by cpufreqd. Moreover the cpupowerd run-script in /etc/init.d just sets cpu frequency governor on the service spawning and stopping. It invokes no persistent daemons to watch for acpi events and temperature sensors.
So, how can cpupowerd be used in place of cpufreqd? |
|
Back to top |
|
|
Logicien Veteran
Joined: 16 Sep 2005 Posts: 1555 Location: Montréal
|
Posted: Wed Sep 11, 2013 5:17 am Post subject: |
|
|
cpufrequtils and cpupower just set the cpu frequency scaling policy on the start and stop of the service. They do not do anything else. They do not have a background daemon. Their configuration file is in /etc/conf.d directory.
cpufreqd is a daemon who can change the cpu frequency scaling policy according to it's configuration file /etc/cpufreqd.conf. cpufrequtils, cpupower and cpufreqd have to be install separately, cpufrequtils and cpupower conflicting each other. _________________ Paul |
|
Back to top |
|
|
ayvango Tux's lil' helper
Joined: 08 Feb 2012 Posts: 118
|
Posted: Wed Sep 11, 2013 5:51 am Post subject: |
|
|
So, should I keep using cpufreqd with newest kernels? |
|
Back to top |
|
|
Logicien Veteran
Joined: 16 Sep 2005 Posts: 1555 Location: Montréal
|
Posted: Wed Sep 11, 2013 7:34 am Post subject: |
|
|
It's depend of what you want. I do not see a big difference between cpupower and cpufrequtils. If you want a fine tune of the CPU according to events you can use cpufreqd. I stop using it since a couple of years. I only switch between ondemand and performance frequency scaling policie from time to time.
For a laptop you can use laptop-mode-tools. Some CPU setup can be done via the ACPI subsystem too. _________________ Paul |
|
Back to top |
|
|
FaeLLe n00b
Joined: 13 Aug 2007 Posts: 13 Location: London
|
Posted: Sun Apr 06, 2014 4:48 pm Post subject: |
|
|
cpufreqd is now masked and will be removed in a month.
Any recommendations on how cpupower can be used a deamon to dynamically scale CPU frequencies? |
|
Back to top |
|
|
Anon-E-moose Watchman
Joined: 23 May 2008 Posts: 6097 Location: Dallas area
|
Posted: Sun Apr 06, 2014 4:54 pm Post subject: |
|
|
I've just started using the freq govs. I installed ondemand, and it seems to be working pretty well.
I have conky set up to monitor each cpu core freq and the Vcore for the cpu.
Ramps up and down smoothly, both for video and compiling, as the freq drops the Vcore power drops correspondingly with it.
Unless you have some special needs I would just use ondemand and see how it works.
As Logicien said compile the performance gov and if needed swap to it. _________________ PRIME x570-pro, 3700x, 6.1 zen kernel
gcc 13, profile 17.0 (custom bare multilib), openrc, wayland |
|
Back to top |
|
|
FaeLLe n00b
Joined: 13 Aug 2007 Posts: 13 Location: London
|
Posted: Sun Apr 06, 2014 6:08 pm Post subject: |
|
|
So if I understand correctly the governor should be able to manage the CPU frequency scaling and there should be no need for any third party tools (unless any special needs exist). Is this correct? |
|
Back to top |
|
|
khayyam Watchman
Joined: 07 Jun 2012 Posts: 6227 Location: Room 101
|
Posted: Sun Apr 06, 2014 9:37 pm Post subject: |
|
|
FaeLLe wrote: | So if I understand correctly the governor should be able to manage the CPU frequency scaling and there should be no need for any third party tools (unless any special needs exist). Is this correct? |
FaeLLe ... correct, infact if you use cpufreqd/cpufrequtils with > 3.7.x you may find the load average becomes somewhat high due (I think) to them being unaware of the changes made when cpupower was introduced.
With cpupower you can still set 'up_threshold', ' sampling_down_factor', 'ignore_nice_load', etc, via the SYSFS_EXTRA variable in /etc/conf.d/cpupower, but there is no facility (as with cpufreqd) to alter the freq based on the application used ... but that shouldn't be necessary anyhow.
best ... khay |
|
Back to top |
|
|
v_2e n00b
Joined: 28 Dec 2008 Posts: 21
|
Posted: Mon Apr 07, 2014 2:50 pm Post subject: |
|
|
Is there a way to perform the CPU governor (or maximum frequency) change depending on the CPU temperature with some tools other than cpufreqd? |
|
Back to top |
|
|
CaptainBlood Advocate
Joined: 24 Jan 2010 Posts: 3595
|
Posted: Tue Apr 08, 2014 3:14 pm Post subject: |
|
|
Hi dear all,
Same question here, as we have 27°C 6:30 AM ...
We're on a rolling out compiling distro, remember?
Not to mention a possibly warming up world?
Thks for your attention.
Last edited by CaptainBlood on Wed Apr 09, 2014 4:44 pm; edited 1 time in total |
|
Back to top |
|
|
khayyam Watchman
Joined: 07 Jun 2012 Posts: 6227 Location: Room 101
|
Posted: Tue Apr 08, 2014 5:53 pm Post subject: |
|
|
v_2e, CaptainBlood ...
It somewhat depends on the cpufreq driver used ... if intel_pstate is used then thermald may work. Also disabling 'turbo' via /sys/devices/system/cpu/intel_pstate/no_turbo will help keep the temp down.
Otherwise, might look under /sys/devices/system/cpu and see what features the driver provides ...
best ... khay |
|
Back to top |
|
|
lexflex Guru
Joined: 05 Mar 2006 Posts: 363 Location: the Netherlands
|
Posted: Tue Apr 08, 2014 6:02 pm Post subject: |
|
|
CaptainBlood wrote: |
Same question here, as we have 27° C 6:30 AM ...
We're on a rolling out compiling distro, remember?
Not to mention a possibly warning up world?
|
Lol misread three times due to the typo, but it seems the warming up is slow enough for any important new packages regarding temperature-monitoring to be developed, introduced in portage, and for any monthly emerge --sync to catch up with the changes |
|
Back to top |
|
|
CaptainBlood Advocate
Joined: 24 Jan 2010 Posts: 3595
|
Posted: Wed Apr 09, 2014 5:13 pm Post subject: |
|
|
Hi dear all,
I've been on cpufreqd for ages now, coz it allows to cap freq according to temp.
And since I moved to where I'm now, I experience shutdown du to overheat detection :
Xen dom0 and un to for domUs are making a fool out of me when competing with my daily simulatenous eix-sync && emerge -uDN world syndroma on my laptop.
khayyam : I thought p-state was obsolete (depreciated). Pardon me if I"m wrong. Anyhow will investigate accordingly. Thks
By the way I run on a Intel Core 2 Duo SU9400 / 1.4 GHz (plz stop laughing). Distcc in configured but untested yet due to lack of HW.
Bye RvryBudy and thanks for your attention. |
|
Back to top |
|
|
khayyam Watchman
Joined: 07 Jun 2012 Posts: 6227 Location: Room 101
|
Posted: Wed Apr 09, 2014 6:37 pm Post subject: |
|
|
CaptainBlood wrote: | I've been on cpufreqd for ages now, coz it allows to cap freq according to temp. |
CaptainBlood ... well, before I switched from cpufreqd to cpupower (this was round about the same time as cpupower was introduced in the kernel) I experienced very weird behaviour with cpufreqd (heavy load ... due I think to it polling parts of cpufreq which had changed) ... this may have been a bug, I can't say, but I stopped using it for that reason. As always, YMMV.
CaptainBlood wrote: | khayyam : I thought p-state was obsolete (depreciated). Pardon me if I"m wrong. Anyhow will investigate accordingly. |
No, "intel_pstate" is that used for processors such as 'Sandy bridge' ...
kernel config wrote: | CONFIG_X86_INTEL_PSTATE: This driver provides a P state for Intel core processors. The driver implements an internal governor and will become the scaling driver and governor for Sandy bridge processors. |
kernel config wrote: | By the way I run on a Intel Core 2 Duo SU9400 / 1.4 GHz (plz stop laughing). |
In which case you're probably using X86_ACPI_CPUFREQ (acpi-cpufreq) which is also a "p-states" driver. I am too (T2500 @ 2.00GHz ... which isn't even a core2), and no thermal control is provided ... you should take a look in /sys/devices/system/cpu/cpufreq/<governor>.
BTW, I'm surprised you can run virtualisation with such a processor ...
best ... khay |
|
Back to top |
|
|
Princess Nell l33t
Joined: 15 Apr 2005 Posts: 916
|
Posted: Wed Apr 09, 2014 10:26 pm Post subject: |
|
|
The whole cpufreqd/cpufrequtils/cpupower situation is quite unsatisfactory for mate users. How can we solve this conundrum?
Code: |
# emerge -upDN world
These are the packages that would be merged, in order:
Calculating dependencies... done!
[ebuild N ] sys-power/cpufrequtils-008-r4 USE="nls -debug"
[ebuild N ] app-text/gnome-doc-utils-0.20.10-r1 PYTHON_TARGETS="python2_7 -python2_6"
[ebuild N ] gnome-extra/yelp-xsl-3.10.1
[ebuild N ] dev-util/gtk-doc-1.20 USE="-debug -doc -emacs -highlight {-test} -vim" PYTHON_SINGLE_TARGET="python2_7 -python2_6" PYTHON_TARGETS="python2_7 -python2_6"
[ebuild N ~] mate-base/mate-applets-1.6.1 USE="networkmanager policykit -ipv6"
[blocks B ] sys-power/cpufrequtils ("sys-power/cpufrequtils" is blocking sys-power/cpupower-3.13)
* Error: The above package list contains packages which cannot be
* installed at the same time on the same system.
(sys-power/cpufrequtils-008-r4::gentoo, ebuild scheduled for merge) pulled in by
sys-power/cpufrequtils required by (mate-base/mate-applets-1.6.1::mate-overlay, ebuild scheduled for merge)
(sys-power/cpupower-3.13::gentoo, installed) pulled in by
sys-power/cpupower required by @selected
For more information about Blocked Packages, please refer to the following
section of the Gentoo Linux x86 Handbook (architecture is irrelevant):
http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?full=1#blocked
#
|
|
|
Back to top |
|
|
Logicien Veteran
Joined: 16 Sep 2005 Posts: 1555 Location: Montréal
|
Posted: Thu Apr 10, 2014 1:35 am Post subject: |
|
|
Cpufrequtils and Cpupower cannot be installed at the same time. You have to choose between one of the two. Only gkrellm-cpufreq depend on Cpufrequtils. So if you do not use this Gkrellm plugin you can switch between Cpupower and Cpufrequtils by unmerge one and emerge the other.
For the moment, I do not see any other package than Cpufreqd who can set the cpu frequency and governor according on the temperature of the processor. It is very usefull with Amd processors to prevent preventive shutdown due to overheating. Maybe laptop-mode-tools can. I haven't use it since long.
Code: | equery d cpufreqd cpufrequtils cpupower
* These packages depend on cpufreqd:
* These packages depend on cpufrequtils:
x11-plugins/gkrellm-cpufreq-0.6.1 (sys-power/cpufrequtils)
* These packages depend on cpupower: |
_________________ Paul |
|
Back to top |
|
|
Princess Nell l33t
Joined: 15 Apr 2005 Posts: 916
|
Posted: Thu Apr 10, 2014 8:36 pm Post subject: |
|
|
The point of my posting was to show that mate-applets has a dependency on cpufrequtils.
Maybe there should be a virtual for it. |
|
Back to top |
|
|
energyman76b Advocate
Joined: 26 Mar 2003 Posts: 2048 Location: Germany
|
Posted: Sun Apr 13, 2014 9:12 pm Post subject: |
|
|
FaeLLe wrote: | So if I understand correctly the governor should be able to manage the CPU frequency scaling and there should be no need for any third party tools (unless any special needs exist). Is this correct? |
correct for many, many years. _________________ Study finds stunning lack of racial, gender, and economic diversity among middle-class white males
I identify as a dirty penismensch. |
|
Back to top |
|
|
netixen n00b
Joined: 26 Sep 2013 Posts: 31
|
Posted: Sun Apr 13, 2014 9:41 pm Post subject: |
|
|
There is a big difference in how the on-demand governor works on different cpus.
For instance some mobile cpus can only step up/down in to 3 frequencies (e.g 1200Mhz 2000Mhz 2200Mhz) while desktop cpus in to more or even step up/down through small steps +/- 10Mhz.
So the on-demand governor can deal with the performance management well but if you have thermal issues you will need to manage it from user-space.
I have written a user-space governor that monitors the thermal zones and adjusts the frequency when trip points are triggered. But I've only tested it on my own machines so it's not ready for the wild.
However if your system is about to catch on fire feel free to give me a shout. |
|
Back to top |
|
|
szatox Advocate
Joined: 27 Aug 2013 Posts: 3129
|
Posted: Mon Apr 14, 2014 9:15 pm Post subject: |
|
|
Some procesors are said to be working better with conservative governor rather than on-demand. I think conservative is better for AMD CPUs and perhaps mobile CPUs. Purpose is the same as on-demand.
Never bothered to verify though, so well, just sain'. |
|
Back to top |
|
|
Anon-E-moose Watchman
Joined: 23 May 2008 Posts: 6097 Location: Dallas area
|
Posted: Mon Apr 14, 2014 10:53 pm Post subject: |
|
|
From linux/Documentation/cpu-freq/governors.txt
The CPUfreq governor "ondemand" sets the CPU depending on the
current usage. To do this the CPU must have the capability to
switch the frequency very quickly.
---
The CPUfreq governor "conservative", much like the "ondemand"
governor, sets the CPU depending on the current usage. It differs in
behaviour in that it gracefully increases and decreases the CPU speed
rather than jumping to max speed the moment there is any load on the
CPU. This behaviour more suitable in a battery powered environment. _________________ PRIME x570-pro, 3700x, 6.1 zen kernel
gcc 13, profile 17.0 (custom bare multilib), openrc, wayland |
|
Back to top |
|
|
defer- Tux's lil' helper
Joined: 11 Jun 2007 Posts: 140 Location: Finland
|
Posted: Mon Apr 14, 2014 11:02 pm Post subject: |
|
|
I have heard that ondemand saves more power than conservative in real word situations. Ondemand finishes tasks more quickly and cpu can go to sleep state more quickly. _________________ https://github.com/defer- |
|
Back to top |
|
|
Gusar Advocate
Joined: 09 Apr 2005 Posts: 2665 Location: Slovenia
|
Posted: Mon Apr 14, 2014 11:18 pm Post subject: |
|
|
defer- wrote: | I have heard that ondemand saves more power than conservative in real word situations. Ondemand finishes tasks more quickly and cpu can go to sleep state more quickly. |
Indeed. That help text for the conservative governor is severely outdated. Race-to-idle is the game nowadays, and conservative completely screws with that. Power saving mostly comes from C-states, not P-states. Even ondemand is outdated in this regard, its constant P-state switching is superfluous. In deep C states, the frequency is 0 anyway, only in the C0 state is the frequency at the one determined by the P-state. That's why Intel developed the intel_pstate driver, which pretty much only switches P-states to handle turbo modes.
Required reading for everyone, it explains in more detail how modern processors work: https://plus.google.com/114657443111661859546/posts/dLn9T4ehywL |
|
Back to top |
|
|
|