Set CPU governor to performance in 18.04

Question

I'm trying to set the CPU governor to performance in 18.04, the problem is that it never survives reboots.

I've tried following these steps in an answer to this question How to set "Performance" instead of "Powersave" as default? but when I reboot the computer the CPU governor switches back to powersaver.

I've also tried creating a file called /etc/rc.local with the line:

cpupower frequency-set --governor performance

That also doesn't survive reboots.

How can I do this?

Solution 1

sudo apt-get install cpufrequtils
echo 'GOVERNOR="performance"' | sudo tee /etc/default/cpufrequtils
sudo systemctl disable ondemand

Solution 2

Short Answer

In /etc/rc.local put in these commands:

sleep 120
cpupower frequency-set --governor performance

---

1 minute after boot automatic switch to Powersave

For whatever reasons my Skylake Intel CPU always starts up in Performance mode and then switches to Powersave mode at the 1 minute mark automatically.

If you set the mode to Performance on startup it will be overridden around the 1 minute Up Time mark to Powersave mode.

In the GIF below, the 3000+ MHz CPU speed at start up appears near the top. The up time appears near the bottom. When up time hits about 1 minute you see CPU MHz drop off. :

---

Program to monitor exact second Powersave invoked

Create this script in /usr/local/bin/watch-gov.sh:

#! /bin/bash

# NAME: watch-gov.sh
# PATH: /usr/local/bin
# DESC: Set governnor to performance and watch for change
#       Ask Ubuntu question: https://askubuntu.com/questions/1021748/set-cpu-governor-to-performance-in-18-04/1084727#1084727
# CALL: called from `/etc/rc.local`
# DATE: Created Oct 18, 2018.

echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
last_gov=$(cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor)
Uptime=$(uptime)
echo "watch-gov.sh: Set to performance at $Uptime " > /tmp/watch-gov.log

for ((i=0; i<300; i++)) ; do
    curr_gov=$(cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor)
    if [ $last_gov != $curr_gov ] ; then
        last_gov=$curr_gov
        Uptime=$(uptime)
        echo "watch-gov.sh: Current governor: $last_gov Uptime: $Uptime" >> \
            /tmp/watch-gov.log
    fi
    sleep 1
done

Call the script in /etc/rc.local before the exit 0 command (explained in detail below).

One minute after logging in look at the output:

$ cat /tmp/watch-gov.log
watch-gov.sh: Set to performance at  17:50:09 up 0 min,  0 users,  load average: 0.00, 0.00, 0.00 
watch-gov.sh: Current governor: powersave Uptime:  17:51:09 up 1 min,  1 user,  load average: 1.89, 0.62, 0.22

Confirmation from this answer states this 1 minute force to powersave governor is controlled by /etc/init.d/ondemand.

---

Sleep 120 seconds before setting Performance Mode

The simplest way to stay in Performance mode is to edit /etc/rc.local and insert these lines before the last line containing exit 0:

sleep 120 # Give CPU startup routines time to settle.
echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

Save the file and reboot.

If you insert the new lines after exit 0 it will never be executed.

To setup /etc/rc.local in 18.04 see: How to Enable /etc/rc.local with Systemd

---

Caveats

Your machine will probably run 10 to 15 degrees C hotter.

You may need to remove other programs that change CPU frequency if they override your Performance settings in /etc/rc.local

Solution 3

Default Ubuntu kernel configurations are such that the performance CPU frequency scaling governor will be used during boot. The relevant section of the kernel configuration file ( /boot/config-4.15.0-36-generic , in this example) is:

#
# CPU Frequency scaling
#
CONFIG_CPU_FREQ=y
CONFIG_CPU_FREQ_GOV_ATTR_SET=y
CONFIG_CPU_FREQ_GOV_COMMON=y
CONFIG_CPU_FREQ_STAT=y
CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE is not set
# CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
# CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set
# CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE is not set
# CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTIL is not set
CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
CONFIG_CPU_FREQ_GOV_POWERSAVE=y
CONFIG_CPU_FREQ_GOV_USERSPACE=y
CONFIG_CPU_FREQ_GOV_ONDEMAND=y
CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y
CONFIG_CPU_FREQ_GOV_SCHEDUTIL=y

#
# CPU frequency scaling drivers
#
CONFIG_X86_INTEL_PSTATE=y
CONFIG_X86_PCC_CPUFREQ=y
CONFIG_X86_ACPI_CPUFREQ=y
CONFIG_X86_ACPI_CPUFREQ_CPB=y
CONFIG_X86_POWERNOW_K8=y
CONFIG_X86_AMD_FREQ_SENSITIVITY=m
CONFIG_X86_SPEEDSTEP_CENTRINO=y
CONFIG_X86_P4_CLOCKMOD=m

But also by default during boot up the ondemand service is executed. It sleeps for 1 minutes and then changes the scaling governor to either interactive, ondemand or powersave, depending on availability. In turn availability depends on which CPU frequency scaling driver you are using. The code is (in multiple locations, search for ondemand):

#! /bin/sh
### BEGIN INIT INFO
# Provides:          ondemand
# Required-Start:    $remote_fs $all
# Required-Stop:
# Default-Start:     2 3 4 5
# Default-Stop:
# Short-Description: Set the CPU Frequency Scaling governor to "ondemand"
### END INIT INFO

# Don't run if we're going to start an Android LXC container:
[ ! -f /etc/init/lxc-android-config.conf ] || exit 0

PATH=/sbin:/usr/sbin:/bin:/usr/bin

. /lib/init/vars.sh
. /lib/lsb/init-functions

AVAILABLE="/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors"
DOWN_FACTOR="/sys/devices/system/cpu/cpufreq/ondemand/sampling_down_factor"

case "$1" in
    start)
        start-stop-daemon --start --background --exec /etc/init.d/ondemand -- background
        ;;
    background)
        sleep 60 # probably enough time for desktop login

        [ -f $AVAILABLE ] || exit 0
        read governors < $AVAILABLE
        case $governors in
                *interactive*)
                        GOVERNOR="interactive"
                        break
                        ;;
                *ondemand*)
                        GOVERNOR="ondemand"
                        case $(uname -m) in
                                ppc64*)
                                        SAMPLING=100
                                ;;
                        esac
                        break
                        ;;
                *powersave*)
                        GOVERNOR="powersave"
                        break
                        ;;
                *)
                        exit 0
                        ;;
        esac

        for CPUFREQ in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
        do
                [ -f $CPUFREQ ] || continue
                echo -n $GOVERNOR > $CPUFREQ
        done
        if [ -n "$SAMPLING" ] && [ -f $DOWN_FACTOR ]; then
                echo -n $SAMPLING > $DOWN_FACTOR
        fi
        ;;
    restart|reload|force-reload)
        echo "Error: argument '$1' not supported" >&2
        exit 3
        ;;
    stop)
        ;;
    *)
        echo "Usage: $0 start|stop" >&2
        exit 3
        ;;
esac

Why is it called "ondemand", but it sets other governors (for example with the intel_pstate driver it will set the powersave governor)? Because this tool pre-dates the intel_pstate driver, back to a time when, by far, the dominant frequency scaling driver was the acpi-cpufreq driver, and "ondemand" was the preferred Ubuntu default governor.

So, one way to boot and stay using the performance CPU frequency scaling governor is to disable the service that changes away from it (which was also mentioned in another answer):

Before:

~$ systemctl list-units --all --type=service | grep ondemand
  ondemand.service                                      loaded    inactive dead    Set the CPU Frequency Scaling governor
~$ cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
ondemand
ondemand

Disable the service:

~$ sudo systemctl disable ondemand
Removed /etc/systemd/system/multi-user.target.wants/ondemand.service.

re-boot, then check again (being sure to wait a minute after the re-boot):

doug@s17:~$ systemctl list-units --all --type=service | grep ondemand
doug@s17:~$
doug@s17:~$ cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
performance
performance

Note: the examples in this answer are from a computer that uses the acpi-cpufreq CPU frequency scaling driver. If you are using the intel_pstate driver, with no ondemand governor, the powersave governor will be used by default.

Anticipated question: Why do my CPU frequencies scale, even when using the performance governor?

Answer: Modern processors scale the CPU frequency, even in performance mode and as a function of the depth of the idle state they go into. If you really want to lock the CPU frequency then disable all idle states deeper than 0. However, note that it will cost a huge huge amount of power.

Personally, and as mentioned in another answer, I use the performance governor or the powersave governor as a function of whatever work I am doing. My scripts are a little different:

$ cat set_cpu_performance
#! /bin/bash
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

for file in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do echo "performance" > $file; done

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

and:

$ cat set_cpu_powersave
#! /bin/bash
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

for file in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do echo "powersave" > $file; done

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

example usage (on a computer using the intel_pstate driver):

$ sudo ./set_cpu_performance
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
performance
performance
performance
performance
performance
performance
performance
performance

Solution 4

What I did was use the file /etc/rc.local

To help you find your paths, use:

find / -name scaling_governor
find / -name scaling_max_freq

This works for my setup, but you just need to edit it for your setup

I added to /etc/rc.local using nano:

# Set CPU Governor and Freq at boot up
 echo "performance" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
 echo 1500000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
 echo 2000000 > /sys/devices/system/cpu/cpu4/cpufreq/scaling_max_freq
 echo "performance" > /sys/devices/system/cpu/cpu4/cpufreq/scaling_governor

directly under the shebang line. Close nano with Ctrl-X and Y to save

Then for 18.04 (may not work on 16.04) on the command line run:

systemctl disable ondemand

Then on the command line - read /etc/rc.local and then reboot:

/etc/rc.local
reboot

if /etc/rc.local chokes and errors then make sure it's chmod +x /etc/rc.local

Solution 5

I am using this bash script to set the performance governor:

#!/bin/bash

sudo cpufreq-set --cpu 0 --governor performance
sudo cpufreq-set --cpu 1 --governor performance
sudo cpufreq-set --cpu 2 --governor performance
sudo cpufreq-set --cpu 3 --governor performance
sudo cpufreq-set --cpu 4 --governor performance
sudo cpufreq-set --cpu 5 --governor performance
sudo cpufreq-set --cpu 6 --governor performance
sudo cpufreq-set --cpu 7 --governor performance

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

Make is executable chmod +x cpu.sh and run at any time You need (ever right after every computer start/reboot).

View more solutions

Comments

[Zanna]

What procedure exactly did you follow? Is the typo in the filename /etc/rc.local only in this question, or did you make that same typo when creating it? This question may not be specific to 18.04 and probably shouldn't be a bug report rather than a question here. I think the question could be considered for reopening if you edit it to add more detail.

[WinEunuuchs2Unix]

The parameter --governor powersave should be --governor performance ...

[Falc]

I followed this command list, but when I reboot the computer the cpu governor switches back to powersaver. I didn't realise there is a typo in the /etc/rc.local command, I got that from another askubuntu post here when I tried to make a script to have the command 'cpupower frequency-set --governor performance' run at startup.

[David Foerster]

Possible duplicate of How to permanently set CPU power management to the powersave governor?

[Falc]

Yeah I've tried following that thread, I get this message: j@j-ncase:~$ sudo update-rc.d ondemand disable update-rc.d: error: cannot find a LSB script for ondemand

[n1ghtm4n4g3r]

@DavidFoerster half-way duplicate... See my comment there with an Ubuntu 18.06 specific update: askubuntu.com/a/1049270/839002 .

[David Foerster]

@Falc: Could you please edit your post when you want to add information? Especially file or program output listings (with the help of the {} button in the editor toolbar) will be much more readable there; alternatively you can use a pastie service for longer listings and include the link of your pastie in your question. Overall it’s best to have everything relevant in one place. Additionally, comments may be deleted for various reasons. Thanks.

[David Foerster]

@Thenightmanager: With your addition the linked question is not a full duplicate. ;-]

[WinEunuuchs2Unix]

@Zanna is correct about reopening after an edit. I believe the question is relevant to all Intel processors that support frequency scaling which probably appeared a decade ago?

[WinEunuuchs2Unix]

Because of your question I updated my answer in your link first link: askubuntu.com/a/936488/307523 to make it work for your question. Thanks.

[j0h]

Not one reference to cgroups (control groups)... looks like i'll have to answer this. cgroups limits cpu usage, and memory use for users in a group.

[WesZ]

One other thing that I do is to edit /etc/init.d/ondemand or ondemand needs to be moved to ondemandzzz to disable it. There is a sleep 60 in /etc/init.d/ondemand before it sets the governor. Even though rc.local will change it to performance at boot time, 60 seconds later it gets changed by /etc/init.d/ondemand. In /etc/init.d/ondemand change the line: echo -n $GOVERNOR > $CPUFREQ to: echo -n "performance" > $CPUFREQ if you decide to keep /etc/init.d/ondemand working...

[Yalok Iy]

Actually, just the last command is enough. After sudo systemctl disable ondemand and reboot: cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor shows performance.

[Csaba Toth]

Which package has cpupower and which GUI you have that animgif from? I'm trying to tune a Ryzen 7 with cpufreq-utils but it feels sluggish even with the performance governor

[WinEunuuchs2Unix]

@CsabaToth I use Conky to display CPU, RAM, HDD, SDD, Network and other system details. You don't need cpupower but it is part of cpufrequtils package. If system is sluggish try Unity desktop instead of default Gnome desktop.

[Csaba Toth]

@WinEunuchs2Unix thanks, I haven't ever heard of Conky, and it was installed onto my system. I'm running Devuan BTW with Xfce4. So Conky says my CPU frequency is 550 MHz (0.55 GHz), which explains the sluggishness and is what I expected, but cpufreq-info tells 3.2 GHz and I'm forcing performance governor. I need to open a Question for this.

[JoelBonetR]

is disable ondemand needed or just recommended?

[doug]

/etc/rc.local & /etc/init.d/ondemand don't exist in 18.04..

[WinEunuuchs2Unix]

@doug To setup /etc.rc.local in 18.04 see: How to Enable /etc/rc.local with Systemd

[yop83]

@JoelBonetR In my case, it was needed. /etc/default/cpufrequtils wasn't enough.

[yop83]

There are so many exotic and complicated solutions out there, but systemctl disable ondemand is by far the simplest option that I've seen anywhere, and it works great. I also had to do echo 'GOVERNOR="performance"' > /etc/default/cpufrequtils to persist between reboots, as mentioned above.

[Willi Mentzel]

instead of modifying each cpu core separately you can specifiy the -r option "modify all hardware-related CPUs at the same time"

[Reza]

@Yalok Iy I tried what you said first but it didn't work for me but the answer did it.

[Yalok Iy]

@Reza are you also using ubuntu 18.04? Perhaps you are using a newer version and the defaults have changed

[Matthias]

Thanks for this great summary. On my machine (Pop_OS 21.04), Kernel 5.13, the ondemand service does not appear to exist? sudo systemctl status ondemand says: Unit ondemand.service could not be found.

[Doug Smythies]

@Matthias : Yes, as far as I know the ondemand service is an Ubuntu thing.

[Matthias]

@DougSmythies I researched it a bit, and it turns out that I was using the intel_pstate scaling driver. It only supports two governors: performance and powersave. I don't think it's Ubuntu related but rather based on the scaling driver. When I switch to intel_cpufreq, I see more, including ondemand. This is a pretty good summary: kernel.org/doc/html/v4.12/admin-guide/pm/cpufreq.html

[Doug Smythies]

You are confusing the ondemand governor with the Ubuntu on demand service. The two are not related at all. The ubuntu service sets the governor after boot. It used to wait one minute, but does't anymore. The service is Named "ondemand" because way back that was all it did, was set the governor to "ondemand", as at the time the intel_pstate driver did not exist. @Matthias .

[Matthias]

Ah yes, makes sense -- thanks for clarifying!