Ryzen CPU power management / scaling on linux (CentOS)?

[unseengundam101]

I am thinking about building a linux Server on Ryzen, probably on CentOS.

I need to how well can linux handle power states and scaling Ryzen CPU frequency on demand. Can it drop Ryzen to low clock speed when idle, handle light and medium load, plus boost to max frequency on high loads.

Looks like AMD Ryzen has 'SenseMI' power management feature set including 'Pure Power', 'Precision Boost', 'Extended Frequency Range (XFR)', 'Smart Prefetch', and "Neural Net Prediction'. Any idea of any or all of these features will work on Linux or CentOS?

 

[whm1974]

What is the current release of CentOS and how recent is it?

 

[mv2devnull]

CentOS 7 is currently based on RHEL 7.5 and the latest kernel update (3.10.0-862.6.3.el7) appeared July 3rd.

I know nothing about Ryzen. CentOS Fora ought to have some users: https://www.centos.org/forums/viewforum.php?f=44

 

[Fallen Kell]

Phoronix attempted to include CentOS 7 in a shootout of linux distributions running on Ryzen last year, but CentOS failed be able to boot. Personally I love RHEL and CentOS, but for new hardware that isn't immediately designed for business use cases, they are slow to get full working support. CentOS 7 is on the 3.x kernel branch, while true ryzen support is really not fully available outside the 4.10 or newer kernels. Yes, Red Hat does backport a lot of the items, but there is only so much that they can do when they are that far behind.

That being said, we should be seeing RHEL 8.x (or whatever they are going to call it) very soon (past history of releases puts the new one out around the time they have released the <Version -1>.5, with RHEL 7.5 out a few months ago.

 

[whm1974]

Phoronix attempted to include CentOS 7 in a shootout of linux distributions running on Ryzen last year, but CentOS failed be able to boot. Personally I love RHEL and CentOS, but for new hardware that isn't immediately designed for business use cases, they are slow to get full working support. CentOS 7 is on the 3.x kernel branch, while true ryzen support is really not fully available outside the 4.10 or newer kernels. Yes, Red Hat does backport a lot of the items, but there is only so much that they can do when they are that far behind.

That being said, we should be seeing RHEL 8.x (or whatever they are going to call it) very soon (past history of releases puts the new one out around the time they have released the <Version -1>.5, with RHEL 7.5 out a few months ago.

I surprised that any distro is still using the 3.x kernel branch, hasn't the 4.x one already been well tested enough?

 

[mv2devnull]

I surprised that any distro is still using the 3.x kernel branch, hasn't the 4.x one already been well tested enough?

RHEL 7 was released June 2014 and had been feature frozen way before that. Hence 3.10.
RHEL 6 was released Nov 2010 and will finally die in 2024. RHEL 6 is on kernel 2.6.32.
https://access.redhat.com/support/policy/updates/errata
https://en.wikipedia.org/wiki/Red_Hat_Enterprise_Linux

The selling point of RHEL (and hence CentOS) is that what you install might run a decade without larger changes.

There must be some pressure on RHEL to get Ryzen supported via backports, or they cannot sell to all server farms (assuming the farms invest in AMD, rather than Intel).

The ELRepo repository does pack "long term" and "mainline" kernels for el-distros:
http://elrepo.org/tiki/tiki-index.php

 

[unseengundam101]

Yeah, CentOs is pretty hold. I am thinking about using Fedora version Red Hat instead now since it more update to date.

 

[ancient pedant]

My 2700x manages its voltages and clockspeeds for all cores quite well on it's own. An idle desktop will see all 16 cores at about 2.6, with one or two cores going up to 4.3 for brief period of time to quickly handle little demands that arise. Heavy loads will see all cores at 3.9 for extended periods. Temperatures as shown by the modified it87driver have gotten up to 74C without throttling, which leads me to believe there might be a 10 degree offset in play.

Sleep states are another matter, and are often blamed for the "random lockup" bug that some ryzen users, including me, experience occasionally. It doesnt correlate with heavy processing demands, which is why some theorize that it has to do with one or more cores being put into a lower C state. I have JUST disable global C states in bios to see if the random, unpredictable bug still happens. Some users do not have this gremlin at all. My core speeds and voltages are still self-regulating quite nicely.

 

[VirtualLarry]

Sleep states are another matter, and are often blamed for the "random lockup" bug that some ryzen users, including me, experience occasionally. It doesnt correlate with heavy processing demands, which is why some theorize that it has to do with one or more cores being put into a lower C state. I have JUST disable global C states in bios to see if the random, unpredictable bug still happens. Some users do not have this gremlin at all

That sounds MIGHTY similar to the "Bay Trail" sleep / C-state related bug or bugs. Maybe they are one and the same?