Hi everyone, I'm quite new to Lemmy (this is actually my fist post!) During the past weeks I've sunk into the magical world of AMD P-States, and, I ended up putting together a quick post that I thought might be useful for someone else.
I'm a Linux amateur, so this could be very much wrong, but I'm very much open to any corrections or improvements :)
1. Requirements
Currently, some of the Zen2 and Zen3 processors support amd-pstate
and the new amd_pstate_epp
scaling driver. You also have to have CPPC support enabled in your UEFI. In the future, it will be supported on more and more AMD processors.
2. amd-pstate vs acpi-cpufreq
There are two methods for adjusting CPU performance on AMD CPU/APUs:
amd-pstate
acpi-cpufreq
acpi-cpufreq
is currently default for most distros, regardless of the CPU in use. on most AMD CPUs this is a limiting factor, as it offers limited performance options with only a few fixed levels for CPU speed.
On newer AMD CPUs and APUs (aka Zen2 and above), there is a more advanced method called Collaborative Processor Performance Control (CPPC mentioned in the requirements), which allows for fine-tuned and continuous adjustments of the CPU frequency, with the potential to provide better performance and energy efficiency compared to the older fixed levels.
And that's where amd-pstate
comes in, as it is a new kernel module that supports the newer and more efficient AMD P-States mechanism.
There are 3 options available, listed below, in order of release:
-
amd_pstate=passive
(Kernel 6.1+) -
amd_pstate=active
(Kernel 6.3+) -
amd_pstate=guided
(kernel 6.4+)
Passive Mode
amd_pstate=passive
When you set amd_pstate=passive
, the processor aims for a certain performance level relative to its maximum capacity. Below a specific point, the performance is average, while above it, the performance remains at its best.
Active Mode
amd_pstate=active
Setting amd_pstate=active
gives low-level control to the processor's firmware. It can prioritize either performance or energy efficiency based on software hints AND the amd_pstate_epp
driver.
The amd_pstate_epp
(Energy Performance Preference) driver provides the firmware with a hint.
On most AMD CPUs, these hints are:
default
performance
balance_performance
balance_power
power
Guided Mode
amd_pstate=guided
Choosing amd_pstate=guided
lets the platform automatically select a suitable performance level within a given range based on the workload.
3a. Configure amd_pstate to either Passive or Guided
To enable the amd_pstate_epp
scaling driver, which also includes instructions for the original amd_pstate
scaling driver, you will need to add a kernel parameter. If you are using PopOS (like me) or any other distribution utilising kernelstub, this process can be easily accomplished with the following steps:
IMPORTANT: The option 'amd_pstate=guided' is only available on Kernel 6.4 or later versions.
- Add the desired kernel parameter by running the following command:
# Add the desired Kernel Parameter
sudo kernelstub -a "amd_pstate=guided" # Change this to passive if preferred
- To confirm that the kernel parameter has been successfully added, use the following command:
# Verify that the kernel parameter has been added
sudo kernelstub -p
Verify amd_pstate
To verify that this is functioning correctly, reboot your machine, and run
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
If amd_pstate
was set to either passive
or guided
, this should now show:
amd-pstate
3b. Configure amd_pstate_epp to Active
To enable the amd_pstate_epp
scaling driver, which also includes instructions for the original amd_pstate
scaling driver, you will need to add a kernel parameter. If you are using PopOS (like me) or any other distribution utilising kernelstub, this process can be easily accomplished with the following steps:
IMPORTANT: The option 'amd_pstate=guided' is only available on Kernel 6.3 or later versions.
- Add the desired kernel parameter by running the following command:
# Add the desired Kernel Parameter
sudo kernelstub -a "amd_pstate=active"
- To confirm that the kernel parameter has been successfully added, use the following command:
# Verify that the kernel parameter has been added
sudo kernelstub -p
Verify amd_pstate
To verify that this is functioning correctly, reboot your machine, and run
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
If amd_pstate
was set to active
, this should now show:
amd-pstate-epp
Configure amd_pstate_epp Energy Performance Preference
The amd_pstate_epp scaling driver introduces a new parameter known as "Energy Performance Preference" (EPP) hint. This setting can be adjusted through sysfs, with two main files controlling it:
-
/sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
: This file displays the current EPP hint for the respective CPU core. -
/sys/devices/system/cpu/cpu*/cpufreq/energy_performance_available_preferences
: This file provides the available EPP hints for the respective CPU core.
To see your current EPP hints (note *
= all CPU cores), use the following command:
cat /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
To view the available EPP hints (which should be the same for all cores), use this command:
cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_available_preferences
# What you see below, is my results on my Ryzen 7 7735HS
default performance balance_performance balance_power power
If you'd like to set the same EPP hint across all cores, for instance, setting EPP to "power" (like in my case), you can use this command:
echo "power" | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
power
NOTE: This is not permament, and will be reverted upon reboot. To make this permanent, you can use multiple tools, or, create a cron job
4. Scaling Driver vs CPU Governor
The Scaling Driver is different than the CPU governor (e.g. powersave
, performance
, ondemand
, schedulutil
, etc.), and the two can be mixed and matched to create your perfect combo.
To check what's the current cpu governor
, use the command below:
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
In my case, that's what I'm seeing:
user@machine ~> cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
powersave
If you've configured amd_pstate=active
, you can mix and match governors with EPP hints. For me the default after I enabled amd_pstate_epp
was governor = powersave
and EPP hint = performance
. Phoronix has an excellent breakdown of all the combinations of governors + EPP hints (referenced in the resources section at the end of this post).
Resources:
-
an amazing Redditor (whose post I cannot find anymore) that served as a basis for this very post (if anyone finds it, please do let me know, and I'll reference them right away)
-
ChatGPT who helped me phrase some sentences a bit better
-
Benchmarks for server using AMD P-State EPP: https://www.phoronix.com/review/linux-63-amd-epyc-epp
-
Benchmarks for Ryzen mobile system using AMD P-State EPP: https://www.phoronix.com/review/amd-pstate-epp-ryzen-mobile
-
Arch Wiki page on CPU Scaling: https://wiki.archlinux.org/title/CPU_frequency_scaling
-
Kernel.org documentation on new AMD P-State driver: https://www.kernel.org/doc/html/latest/admin-guide/pm/amd-pstate.html
Eeeeh, penzaaaavi, voleeeeevi, guarda li che facciah, non se lo aspettavah! 😂
Glad it's been of help :)