AMD P-State Core Performance Boost To Be Merged For Linux 6.11

[phoronix]

Phoronix: AMD P-State Core Performance Boost To Be Merged For Linux 6.11

Linux 6.11 is shaping up to be an exciting summertime kernel cycle for AMD Ryzen owners. The newest feature now being queued ahead of next month's merge window is Core Performance Boost support within the AMD P-State CPU frequency scaling driver...

https://www.phoronix.com/news/AMD-Core-Perf-Boost-Linux-6.11

[pWe00Iri3e7Z9lHOX2Qx]

I got exited about this, then I realized the preferred core stuff may be a different set of patches.

[MastaG]

I got exited about this, then I realized the preferred core stuff may be a different set of patches.

Preferred core and zen2 threadripper support has already landed for 6.10.
Afaik all of the other patchsets are now queued for 6.11.
This means that 6.11 will be an exciting kernel for all zen2 and newer users

[ssokolow]

Always nice to have more options but, at least at 65W TDP, adding a Noctua cooler works so well for just bringing every core up, and the dGPU is a larger-enough source of heat that I don't think I'll need to micromanage my boost states.

[pWe00Iri3e7Z9lHOX2Qx]

Preferred core and zen2 threadripper support has already landed for 6.10.
Afaik all of the other patchsets are now queued for 6.11.
This means that 6.11 will be an exciting kernel for all zen2 and newer users

Super cool, thanks for the heads up! I've been eagerly awaiting that one for my laptop and desktop. With those in place and sched_ext possibly landing in 6.11 it's going to be fun times doing performance testing in games with LAVD.

[skeevy420]

Always nice to have more options but, at least at 65W TDP, adding a Noctua cooler works so well for just bringing every core up, and the dGPU is a larger-enough source of heat that I don't think I'll need to micromanage my boost states.

Heh. That's what I did, too. That and lots of fans correctly placed. The big ole CPU cooler can't do its thing if it doesn't have air coming and going.

But I alternate between Performance cranked to 11 and letting the system do its thing erred towards energy savings depending on what I'm doing. Firefox and Phoronix posting doesn't need it cranked to 11. Games and compilers do. Feral's Gamemode is pretty handy for that.

I wouldn't have been killed if P State didn't lower the CPU because I was camping in the corner looking left and right. Cheating ass kernel governors

[openminded]

This is great! Now the only thing that I need to figure out is how to switch the boost off dynamically for processes of specific type and switch it back on when such process exits or changes core it's running on.

[ssokolow]

Heh. That's what I did, too. That and lots of fans correctly placed. The big ole CPU cooler can't do it's thing if it doesn't have air coming and going.

When I upgraded off my 2011 PC in an Antec P182 case, (I kept the original complete and usable until after I shook the bugs out of the new one) I didn't like my options for cases so I built the new one in my younger brother's old Antec P183.

Granted, it's still got the original fans but, when I pull the NVMe that just reported some bad sectors in a Firefox site storage SQLite file last night and RMA it, I'll probably swap in the Noctua push-pull configuration I put in my 2011 PC a couple of years ago at the same time as reinstalling from Kubuntu 22.04 LTS to 24.04 LTS, buying another NVMe, and researching the least error-prone way to set up booting from a ZFS mirrored pair to get rid of that source of potential downtime or data loss.

(Drive failures aside, my bulk drives are a ZFS mirror that already does hourly/daily/monthly snapshotting using sanoid and it'd be nice to extend that bit of PEBKAC-proofing to my home directory, which is currently on ext4 with nightly rsync to the bulk drives before the whole lot gets rsync'd to the USB3 external backup drive.)

...oh, and I should also take the opportunity to swipe the pair of 10TB bulk drives from the 2011 machine while I'm at it and reinstate a "tier 3" that doesn't get automatic external backup and is meant for stuff I have manual DVD+R-with-dvdisaster backups of like my installed GOG games. (Yeah. I'm a bit of a data-hoarder. Even without the 10TB drives from the old one, at least half the cost of this build went to storage.)

...but yeah, good case for cooling if you're the kind of person who rules out modern cases for being "not enough 5.25" drive bays or not enough cooling or not enough noise-mitigation measures". I still like the aesthetics of the P182 more though, so I'll probably build my next PC in that one and just use the 3.5" drive bay to get my USB3 ports, like I did for USB Type-C on the P183.

I wouldn't have been killed if P State didn't lower the CPU because I was camping in the corner looking left and right. Cheating ass kernel governors

*chuckle* I generally game on a dedicated Win7 gaming rig (made out of a hand-me-down HP prebuilt, a hand-me-down GPU, a 500GB SSD I wound up not using for what I bought it for, and a PSU I bought as an emergency spare) that's KVMed together with this... partly so I don't have to futz around with Wine or Proton for the Windows-only stuff but also partly because it's fundamentally more reliable.

(eg. In non-mouselook games, I can use Barrier plus the monitor's input switcher instead of the hardware KVM and then seamlessly reach over and pause YouTube on one of my wing monitors or scroll a manual PDF on another monitor without fear that the game will do something like minimizing from loss of focus or pausing despite my choice of interaction device being a gamepad... there's something suited to Wayland's "implement this in one central place and debug the crap out of the edge cases" attitude... don't let windows know if they have focus and provide a dedicated Android-style suspending system.)

[Anux]

This is great! Now the only thing that I need to figure out is how to switch the boost off dynamically for processes of specific type and switch it back on when such process exits or changes core it's running on.

This should already be doable with the amd_pstate guided mode and a little script? Just supply your governor with a max frequency and your ready to go.

[openminded]

This should already be doable with the amd_pstate guided mode and a little script? Just supply your governor with a max frequency and your ready to go.

I already doing that with cpupower's max frequency settings for all cores. But since only SSL related tasks (e.g. chrome, git-clone-https) fail, I would like to limit clocks on cores which those processes are running on, and it would be great if that could be automated.