Ryzen CPPC and Garuda Linux

Hi there,

So I have seen this article on Phoronix : Updated AMD P-State Driver Posted For Improving Linux Power Efficiency - Phoronix

and I was interested to know how the Garuda Team will implement this. This is basically being pushed by AMD and Valve to give Ryzen the same power saving and performance benefits that they enjoy on windows. It is roughly equivalent to Ryzen Balanced power profile on Windows which is how AMD intended the CPU to work. Currently Ryzen is has a non optimal implementation on GNU/Linux in how it runs which is why we are using the performance governor as a stopgap measure.

As I am using Ryzen here myself and Garuda Dragonized I know that currently the performance tweaks script changes the CPU governor to performance mode rather than default schedutil.

Since the new Ryzen CPPC control for the CPU clocks, will require schedutil, will Garuda Linux switch Ryzen CPU's to schedutil in the performance tweaks script?

Reason I ask is that currently under performance governor, the CPU is basically running full clocks and hot so it would be nice if this gets mainlined by Linus Torvalds in a future kernel like maybe 5.17. Then at that point perhaps the performance script in garuda assistant may need to be modified in respect to Ryzen CPUs to enable the schedutil governor and thus the AMD CPPC.

Many Thanks

1 Like

Please, post your inxi -Faz. :slight_smile:


We do not create a custom kernel, which is why this request is misdirected here :eyes: it should go to the Linux-zen kernel repo (which we use) instead.
I think @ptr1337 was planning to add, or added this already to his kernel builds tho :thinking:


Yes. If the performance governor is a workaround, and schedutil is better in a future kernel (5.17 or otherwise), then the defaults can be changed. The kernel changes have to be implemented before they can be used.



ive already merged the patch into the kernels, you can switch your governor easily with cpu power btw.

It does not require schedutil but has probally the best power/watt value.
Ondemand is running as the best for me.

The new amd CPPC disables sometimes the pstate driver, you can change the behavior with adding following into the cmdline:

And trust me, AMD kernels are running much better then the intel pstate driver :stuck_out_tongue:

Here you will find all schedulers, with a patchset. You can choose in the PKGBUILD some toggles, for example enabling llvm/lto.



all working here with amd-pstate.

using linux-xanmod-tt-5.13.3 ( because of bcache in 5.15 kernels below 5.15.3 )

its a treat to see the cpus running at 550 MHz :slight_smile:

I had to do a bios update though... but now bios has cppc support and enabled. ( at least i never saw it there before doing the bios update )

before the bios update, cpupower could not switch to ondemand governor and would run at 2200 MHz minimum.

Fun fact: I just found this topic. And was actually searching for a github comment that mentioned the shared_mem kernel command line. Just though of fixing this now because of trying 5.15.3 with bcache, and xanmod enable amd pstate.

EDIT: would provide a screenshot, but spectacle is being a lousy spectator and crashing right away.
EDIT2: I didn't enable "amd_pstate.shared_mem=1"

That's fine for people who have systems that receive BIOS updates... :sob:


Didn't I only received bios updates, but I had several of them to do!!!!

was running on F51, updated to F62d.

bought computer with F50.... updating to F51 fixed some things right away... don't remember exactly what it was ... but was something major! ( like big ... really big.... really must have thing, if I remember correcly, related to AMD AGESA )

EDIT: I hate doing bios updates because when I was a kid I sent a motherboard to heaven with a bios update.
My old faithfull trusty i3770k almost did the same thing... but managed to keep it in this world ( discovered that it would boot to an old bios in a usb pen, if it couldn't boot )
Even now... even though MB has a backup bios, my heart was pouncing while doing it.


This topic was automatically closed 2 days after the last reply. New replies are no longer allowed.