Arch Linux

fixide

Member

Registered: 2013-04-27

Posts: 44

Re: CPU governor Arch Linux Vs Ubuntu

./stats.sh mylog.txt
median: 3500.13
count : 388

Original kernel

Edit : Custom

zgrep CONFIG_HZ /proc/config.gz | sed '/^#/d' && zgrep NO_HZ /proc/config.gz
CONFIG_HZ_1000=y
CONFIG_HZ=1000
CONFIG_NO_HZ_COMMON=y
CONFIG_NO_HZ_IDLE=y
# CONFIG_NO_HZ_FULL is not set
# CONFIG_NO_HZ is not set

./stats.sh mylog.txt
median: 925.262
count : 512

A real difference!

Why tickrate is not set to 1000 by default?

Last edited by fixide (2015-11-14 11:34:47)

Arup

Member

From: Earth

Registered: 2014-02-20

Posts: 95

Re: CPU governor Arch Linux Vs Ubuntu

In 2012 there was a discussion in Arch general mailing list about a guy who was trying to put a co-relation between CONFIG_HZ=300 and systemd, his whole point was a conjecture but not based on science but a probability and therefore he was not entertained much. Also if one does a google search, there are various threads regarding the 300MHz config for Arch kernel past and present including one at nvidia forum. I think this issue needs further looking into by developers. https://www.mail-archive.com/arch-gener … 29815.html

Arup

Member

From: Earth

Registered: 2014-02-20

Posts: 95

Re: CPU governor Arch Linux Vs Ubuntu

Have been doing some extensive testing with Ubuntu 14.04.3 which uses 3.19 kernel and Arch latest with all updates. The cpu-freq indicator in Arch shows near about turbo for my i7 4790 and yet the temperatures whether under full load or idle compare similarly to Ubuntu even though the later does the frequency modulation under idle which Arch doesn't. I haven't measured power yet but with all things being equal, ideally the high frequency should also make the CPU run hotter than Ubuntu's lower frequency at idle and yet both temps are same.

graysky

Wiki Maintainer

From: :wq

Registered: 2008-12-01

Posts: 10,600

Website

Re: CPU governor Arch Linux Vs Ubuntu

./stats.sh mylog.txt
median: 3500.13
count : 388

Original kernel

Edit : Custom

zgrep CONFIG_HZ /proc/config.gz | sed '/^#/d' && zgrep NO_HZ /proc/config.gz
CONFIG_HZ_1000=y
CONFIG_HZ=1000
CONFIG_NO_HZ_COMMON=y
CONFIG_NO_HZ_IDLE=y
# CONFIG_NO_HZ_FULL is not set
# CONFIG_NO_HZ is not set

./stats.sh mylog.txt
median: 925.262
count : 512

A real difference!

Why tickrate is not set to 1000 by default?

Told ya

---

CPU-optimized Linux-ck packages @ Repo-ck  • AUR packages • Zsh and other configs

fixide

Member

Registered: 2013-04-27

Posts: 44

Re: CPU governor Arch Linux Vs Ubuntu

Have been doing some extensive testing with Ubuntu 14.04.3 which uses 3.19 kernel and Arch latest with all updates. The cpu-freq indicator in Arch shows near about turbo for my i7 4790 and yet the temperatures whether under full load or idle compare similarly to Ubuntu even though the later does the frequency modulation under idle which Arch doesn't. I haven't measured power yet but with all things being equal, ideally the high frequency should also make the CPU run hotter than Ubuntu's lower frequency at idle and yet both temps are same.

I do not think that the temperature is a reliable indicator to justify a difference in consumption when the load is idle. You can have a difference of 10W without temperature difference when ventilation is sufficient.
It is fully loaded as there are differences of temperature at different frequency...

I can underclock my cpu @ 1ghz and a few volts unless I always have exactly the same temperature in idle. Same if I overclock to 4.5ghz. 27-30° in all cases with the same fan speed (but i have a thermalright macho as heat sink..). We need a real power mesure !

I'm on linux-CK (from greysky's repos) with 1ghz tickrate now . Linux Mithril 4.3.0-1-ck #1 SMP PREEMPT Sun Nov 15 13:47:14 EST 2015 x86_64 GNU/Linux
model name      : Intel(R) Core(TM) i7-4770K CPU @ 3.50GHz
cpu MHz         : 846.562

Last edited by fixide (2015-11-22 13:19:01)

ewaller

Administrator

From: Pasadena, CA

Registered: 2009-07-13

Posts: 19,808

Re: CPU governor Arch Linux Vs Ubuntu

I do not think that the temperature is a reliable indicator to justify a difference in consumption when the load is idle. You can have a difference of 10W without temperature difference when ventilation is sufficient.

Where are you measuring that temperature? If you are talking about a sensor on the fins of the heat sink, maybe I could see this.  If you are talking about junction temperature, no way.  That is not consistent with the laws of thermodynamics.   
Heat will only flow between two points if there is a temperature differential.  If you are putting in an additional 10W, you have to take out an additional 10W.  Inside a silicon chip, the only cooling process is available is conduction.  The heat flows either out the balls of the BGA package into the circuit board, or to the case of the package where there is an lossy interface to the heat sink.   If you can keep the heat sink at a constant temperature, that is a great design.  But, 10W more in the chip still means that (at equilibrium) you have to increase the heat flow into the heat sink by 10W.  The ways you do that are to decrease the thermal resistance, or to increase the temperature differential.    Now, there are some bi-phase heat sink interface material out there that have better wetting when they liquify, I don't think that is what is happening here.  My point is,  if you are putting in 10 more Watts, and your "cold" reservoir stays at the same temperature, the temperature in the junctions will increase.

tl;dr  Either you are measuring the temperature at the heat sink, or your instrumentation is lying to you.

Last edited by ewaller (2015-11-23 00:37:02)

---

Nothing is too wonderful to be true, if it be consistent with the laws of nature -- Michael Faraday
Sometimes it is the people no one can imagine anything of who do the things no one can imagine. -- Alan Turing
---
How to Ask Questions the Smart Way

Arup

Member

From: Earth

Registered: 2014-02-20

Posts: 95

Re: CPU governor Arch Linux Vs Ubuntu

The RPM of my Evo 212 CPU cooler also stays the same along with temp for my i7 on both Arch and Ubuntu but yes, for concrete conclusion, need to attach a watt meter. elaller above makes a valid point, if the CPUs are using more power, there has to be more heat as well.

Last edited by Arup (2015-11-23 00:40:09)