-
Notifications
You must be signed in to change notification settings - Fork 2
Open
Description
AMDs newer processors have a per core RAPL. (Source: https://www.phoronix.com/news/AMD-Linux-PMU-Power-Per-Core)
If we pin a process to a core like stress-ng and forbid that other processes get scheduled we can derive the coefficients in an exemplary and verifyable case.
I can find it in the newest AMD developer documentation and since the kernel patch was tried already in 2023 I believe the feature is around for at least 2 years. Exact date is however not given in the documentation.
Only thing we know so far it lives in 56h
I will fire up my AMD machine in the next weeks and check if I can ready via perf events or via MSR

Metadata
Metadata
Assignees
Labels
No labels