September 1, 2018 · linux howto archlinux

Improving Linux battery life and thermal efficiency

I recently switched from MacOS to Linux for my personal/work machine. As much as I like my new laptop I really missed all the battery life that my previous Macbook Pro was capable of.

I decided to do something about it and improve battery life of my setup as much as I could. This post is part me saving it for future reference and part sharing it with you, the curious reader.


Setup

I have a System76's Oryx Pro 4 (2018) with the following characteristics:

And I run the following software (at the time of this writing):

------------- 
OS: Arch Linux x86_64 
Host: Oryx Pro oryp4 
Kernel: 4.18.5-arch1-1-ARCH  
Packages: 900 (pacman) 
Shell: zsh 5.5.1 
WM: i3 
------------------

Pop OS!

System76 has their own Linux distribution Pop! OS. By default uses Gnome and comes preloaded with some good stuff and uses an older version of the kernel. It must be the LTS kernel since they have to support it and it would be crazy to support every version out there.

When I tested this machine with Pop OS! I got a discharge rate of about ~11-10W idle, just powertop running. I do not remember if I had the keyboard backlight on and what the screen backlight level was so it may not be a fair comparison but it's a good baseline.

Archlinux

I use Archlinux and i3-gaps which are way lighter than Ubuntu (Pop! OS base) and Gnome. With my current setup I get ~8.5 W discharge rate while idling, just running powertop.

This is before I run anything related to power savings so it comes to show how much Linux has improved.


Power savings configuration

Kernel command line

I boot the kernel adding the following to the command line:

nowatchdog

The interesting bit for our current discussion is the nowatchdog paramater which turns off lockup detectors in the kernel which saves a tiny fraction of power and performance.

Kernel module configuration

I have the following kernel module configuration file:

# /etc/modprobe.d/power-management.conf
# Wifi power saving
options iwlwifi power_save=1 d0i3_disable=0 uapsd_disable=0
options iwldvm force_cam=0

# USB autosuspend
options usbcore autosuspend=5

# Sound card power save
options snd_hda_intel power_save=1

These are here for reference only and to show the concept of doing some research on what drivers your machine is using and what do they need to have increased power savings. You don't want to copy these as-is because it may turn your machine unusable.

TLP – Linux Advanced Power Management

From their project page:

TLP brings you the benefits of advanced power management for Linux without the need to understand every technical detail. TLP comes with a default configuration already optimized for battery life, so you may just install and forget it. Nevertheless TLP is highly customizable to fulfil your specific requirements.

TLP is a daemon that turns on power saving flags in hardware, configures the CPU, configures kernel parameters, etc, that are good for power savings. As the project says, the defaults are pretty good but you can always tweak it a little bit more.

The changes I make to the default configuration are:

DISK_IDLE_SECS_ON_BAT=5
#DISK_DEVICES="sda sdb"
#DISK_APM_LEVEL_ON_AC="254 254"
#DISK_APM_LEVEL_ON_BAT="128 128"
DISK_IOSCHED="keep"

There is some fine tweaking possible for newer CPUs with TLP to manage cpu power hints, pstates, turbo settings, etc. I don't use them since I run System76's system76-power(github) to handle the CPU profiles because it is easier.

Archlinux
sudo pacman -S tlp
sudo systemctl enable --now tlp.service
sudo tlp-stat

fancontrol

fancontrol is a daemon, that as its name says, controls fans in your computer. That is, if the computer maker decided to give you that power. In Linux, such control is done through the PWM (Pulse Width Modulator) interface.

PWM allows you to control the power that goes into an electrical line (I am being intentionally simplistic here). If you have fan PWM lines available you can control how much power goes into them, and in turn, how fast they spin.

fancontrol is part of the lm_sensors package in archlinux and ubuntu. fancontrol is a tool to control a fan speed based on some temperature thresholds. This is useful because usually the auto control in the fans is overly aggressive, turning fans on very rapidly when the CPU activity increases a tiny bit. And, the faster a fan spins, the more power it consumes.

You have to take several things into account here, the temperature limits of the hardware you want to control the fan of, the ambient temperature of where you live. Say, if the CPU wants to be at 30C by default but your ambient temperature is 40C those fans are going to spin forever. You want to be careful with this. You most likely don't want to fry your electronics.

If you install the lm_sensors package and run the sensors utility you can see which hardware temperature monitors you have available and what are their temperature limits. Plus, if you have fans around.

Mine looks like this:

$ sensors
coretemp-isa-0000
Adapter: ISA adapter
Package id 0:  +39.0°C  (high = +100.0°C, crit = +100.0°C)
Core 0:        +37.0°C  (high = +100.0°C, crit = +100.0°C)
Core 1:        +38.0°C  (high = +100.0°C, crit = +100.0°C)
Core 2:        +37.0°C  (high = +100.0°C, crit = +100.0°C)
Core 3:        +38.0°C  (high = +100.0°C, crit = +100.0°C)
Core 4:        +36.0°C  (high = +100.0°C, crit = +100.0°C)
Core 5:        +37.0°C  (high = +100.0°C, crit = +100.0°C)

system76-isa-0000
Adapter: ISA adapter
CPU fan:         2533 RPM
GPU fan:         4083 RPM
CPU temperature:  +38.0°C
GPU temperature:   +0.0°C

Now, that is one I care about, the system76-isa-0000 has a fan for the CPU, and turns out it is very aggressive and since I live in a very warm place to begin with, I want to tone it down so I can save a bit of juice.

Run the pwmconfig utility which will create an /etc/fancontrol file which looks like this:

# Configuration file generated by pwmconfig, changes will be lost
INTERVAL=10
DEVPATH=hwmon1=devices/platform/system76
DEVNAME=hwmon1=system76
FCTEMPS= hwmon1/pwm1=hwmon1/temp1_input
FCFANS= hwmon1/pwm1=hwmon1/fan1_input
MINTEMP= hwmon1/pwm1=40
MAXTEMP= hwmon1/pwm1=75
MINSTART= hwmon1/pwm1=14
MINSTOP= hwmon1/pwm1=12

(Don't copy this. This is very hardware dependant, run pwmconfig)

Enable the fancontrol.service and voila.

Archlinux
sudo pacman -S lm_sensors
sudo pwmconfig
sudo systemctl enable --now fancontrol.service

Fans are now under control.

thermald

Linux thermal daemon (thermald) monitors and controls temperature in laptops, tablets PC with the latest Intel sandy bridge and latest Intel CPU releases. Once the system temperature reaches a certain threshold, the Linux daemon activates various cooling methods to try to cool the system.

Now, the difference between the fans and thermald is that thermald uses passive cooling methods built into these CPUs.

thermald is a last barrier. When the CPU is under heavy load and not even the fan can keep it cool thermald uses several passive cooling features built into newer cpus (Intel's, not sure about AMD) to minimize power and temperature.

This is my current configuration:

/etc/thermald/thermal-conf.xml
<?xml version="1.0"?>
<ThermalConfiguration>
  <Platform>
    <Name>Override CPU Passive</Name>
    <ProductName>*</ProductName>
    <Preference>QUIET</Preference>
    <ThermalZones>
      <ThermalZone>
        <Type>cpu</Type>
        <TripPoints>
          <TripPoint>
            <Temperature>80000</Temperature>
            <type>passive</type>
            <ControlType>SEQUENTIAL</ControlType>
          </TripPoint>
        </TripPoints>
      </ThermalZone>
    </ThermalZones>
  </Platform>
</ThermalConfiguration>

Which pretty much means that if the CPU goes higher than 82C it should use all matters of passive cooling at its disposal to cool it down. I've tested this and it's amazing how even using all the cores heavily it stays at 82C.

After you configure it you just enable the daemon and that's it. It will do it's thing when it must.

Archlinux
sudo pacman -S thermald
sudo systemctl enable --now thermald.service

Undervolting

Note: This is experimental.

Undervolting is similar to overclocking, except instead of increasing speed, you are trying to keep your current speed by using less voltage.

This works because due to tiny differences in manufacturing processes CPU voltage and power specifications are not the same between units. They have ranges and tolerances. This is what lets you overclock and undervolt.

When a CPU maker builds a CPU they make sure to feed it voltage which can guarantee they meet the specified clock speed. But it is very likely any given CPU can work with less voltage and still work within the specified speeds and stability.

What this accomplishes for us is using less power to feed the CPU and that saves juice. Less voltage also mean less heat, which in turn means less fan noise and spinning, which also saves juice.

I use intel-undervolt(github) to set the voltages, but there are other scripts and tools available.

You have to find the specific voltages and tolerances that your CPU can work with, you do this by decreasing the voltage just a tiny bit, do some heavy stress testing for a good chunk of time. If it works, you decrease it a bit more, and you continue doing it till you find a set of voltages that work for your machine.

This is how my (current) /etc/intel-undervolt.conf looks like:

# CPU Undervolting
# Usage: apply ${index} ${display_name} ${undervolt_value}
# Example: apply 2 'CPU Cache' -25.84

apply 0 'CPU' -160
#apply 1 'GPU' 0
apply 2 'CPU Cache' -160
#apply 3 'System Agent' 0
#apply 4 'Analog I/O' 0

# TDP Alteration
# Usage: tdp ${short_term} ${long_term}
# Usage: tdp ${short_term}/${time_window} ${long_term}/${time_window}
# Example: tdp 45/0.002 35/28

# Critical Temperature Offset Alteration
# Usage: tjoffset ${temperature_offset}
# Example: tjoffset -20

# Daemon Update Inverval
# Usage: interval ${interval_in_milliseconds}

interval 5000

The reason the GPU is commented out is because apparently it is already undervolted by quite a bit by default. I recommend you use sudo intel-undervolt read to check your default values. You may find something you don't want to change.

Archlinux
git clone https://aur.archlinux.org/intel-undervolt.git ; cd intel-undervolt
makepkg -sci
sudo systemctl enable --now intel-undervolt.service

We're loading the CPU with mprime to test thermald and intel-undervolt. thermald brings the temp down to 82C and intel-undervolt brings the temp down when we reduce the voltage fed.

Results

With all of this configuration, my idle power usage doesn't really change very much. It starts showing when you actually put some load and start using the system. My overall battery life went from about ~4 hours to about ~5.5 hours. Which is pretty good. It still sucks compared with the MBP but this machine is a power house. I didn't really expect to get 10 hours of battery with it.

Thermals improved quite a bit too, and that's pretty important to me because I live in a very hot place.

That is it. If you have any suggestions how to improve this guide, or how to increase battery life please do share!


Changelog

A lot of people have made suggestion. I have changed my configurations based on them and the post too. I thought it would be a good idea to keep track of them.