Announcement

Collapse
No announcement yet.

Intel Sends Out Linux Patches For Speed Select Core-Power Controls

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel Sends Out Linux Patches For Speed Select Core-Power Controls

    Phoronix: Intel Sends Out Linux Patches For Speed Select Core-Power Controls

    Coming to Linux last year with the 5.3 kernel was Intel Speed Select Technology support as a Cascade Lake feature for optimizing the per-core performance configurations to favor certain cores at the cost of reducing the performance capacity for other CPU cores. That Intel Speed Select (SST) support for Linux is now being enhanced with core-power controls...

    http://www.phoronix.com/scan.php?pag...ect-Core-Power

  • #2
    Optimizing the per-core performance configurations to favor certain cores at the cost of reducing the performance capacity for other CPU cores.
    Is there a task scheduler to take advantage of that? I remember ARM getting a big.little one to bias assignment based on workload of tasks to the more powerful/performant cores vs the lower power cores for lighter / multi-threaded work, something like that. Or do existing schedulers take note of a long running task putting a full load on thread/core and it gets migrated to others for thermal/boost reasons?(at expense of what I think is referred to as context switch or losing it's cache which could harm performance..?)

    Maybe I'm talking non-sense and it doesn't matter. I remember back in 2017 with Ryzen being a new thing and the affordable high core count with ThreadRipper(EPYC was sold out or priced much higher at the time due to demand), but they couldn't hit the same frequency Intel chips were offering which was really useful for single threaded workloads. Company I was at had a software that had several processing stages where some where single threaded restrained for a long time, I thought a desktop equivalent of the ARM big.little cores would be neat, but others said that's what desktop CPUs cater to with their turbo/boost frequencies, having "big" or "little" cores wouldn't achieve anything?

    That does seem to make sense now with turbo/boost frequencies scaling based on how much load the CPU is under, so that a single core/thread can go over 5Ghz.

    Comment

    Working...
    X