Jtdcftul

I run distributed computing projects, which typically want to use 100% of the CPU. How do I limit the CPU usage in terms of temperature instead of percent usage? Also, what is the maximum safe temperature to keep an Intel i5 running 24/7? (With no CPU limit FahCore_a4 causes this machine to run at 82 degrees Celsius.)

On this webpage there is a bash script that will attempt to keep your CPU below a specified temperature. http://seperohacker.blogspot.com/2012/10/linux-keep-your-cpu-cool-with-frequency.html

You just need to provide it with your desired maximum temperature, and it will throttle your CPU(s) in an effort to stay below that temperature.

Shameless plug- I wrote and maintain the above script.

Here is how I solved it using bash. If anyone comes up with a better daemon (better at staying near the target temperature) please post it.

#!/bin/bash



while true; do

        val=$(sensors | awk '/Core 0/ {print $3}')

        max="+60.0"

        if [[ "$val" < "$max" ]]

        then

                killall cpulimit

                sleep .1

        else

                cpulimit -e FahCore_a4 -l 99 &

                sleep 1

        fi

        clear

        sensors

done

The CPU itsself has a mechanism where it powers itsself down if it gets too hot. (maybe not if you disable SMI interrupts, I'm not sure about that.)

The main user-space application is the lm-sensors package. After installing it run sensors-detect to set it up, if your machines are the same you can run this once, and use the resulting findings everywhere.

CPU frequency is easily controled via the cpufreq driver subsystem. see https://wiki.archlinux.org/index.php/CPU_Frequency_Scaling

You could write a daemon that uses lm-sensors to poll the temp and if its too hot turn down the cpu frequency.

This is how I solved my overheating problems which were not caused by a specific process, but by failure of my laptop to dissipate the heat caused by running on a high CPU load for longer periods of time. The main difference to the solution offered by da code monkey himself is that I use cpu frequency scaling, instead of using cpulimit on a specific process. Since I have Psensor installed and running anyway, I piggyback on Psensor. A daemon like da code monkey suggested should also work, however two thresholds (min and max) are required. In Psensor (Sensor Preferences->Alarm) I set an alarm for the relevant sensor if becomes higher than the high threshold (I use 85) or lower than the low threshold (I use 80). In Preferences->Sensors->Script executed when an alarm is raised I call my script heatcontrol.sh like this:

/ALLUSER/heatcontrol.sh powersave ondemand 82

The first parameter is a valid scaling_governor which reduces CPU frequency, e.g. powersave. The second parameter is the default scaling_governor - ondemand for most systems. The third parameter is a temperature between low and high threshold (the exact value is not important). Psensor adds two more parameters: a sensor id and the temperature reported.

And this is my heatcontrol.sh script:

#!/bin/bash

 # TempNZ is temperature reported by Psensor; strip of trailing °C

TempNZ=$5 

bnum=`expr index "$TempNZ" "°"`

if (( bnum < 2 )); then

    echo "ERROR"

fi

TempNZ=${TempNZ:0:$(($bnum-1))}

STR=$1

# Is this a low threshold alarm?

if [ "$TempNZ" -le "$3" ]; then

    STR=$2

fi

sudo sh -c "echo '$STR' > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor"

# echo new scaling_governor

sudo cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

# play sound - audio feedback - optional

paplay /usr/share/sounds/ubuntu/stereo/message-new-instant.ogg

Works fine for me (UBUNTU 14.04 LTS).

Being a novice to Linux and bash, I used a number of resources, including:

https://wiki.archlinux.org/index.php/CPU_frequency_scaling

http://elinux.org/Jetson/Performance#Viewing_the_current_CPU_status