• P cores and E cores

    From Andy Burns@usenet@andyburns.uk to alt.comp.os.windows-11 on Mon Nov 24 17:50:52 2025
    From Newsgroup: alt.comp.os.windows-11

    On a Win11 machine with an i5-1335U (2x P cores with hyperthreading, and
    8x E cores without HT)

    For manually setting CPU affinity, is it guaranteed that CPU0 to CPU3
    are the P cores and CPU4 to CPU11 are E cores?


    --- Synchronet 3.21a-Linux NewsLink 1.2
  • From Paul@nospam@needed.invalid to alt.comp.os.windows-11 on Tue Nov 25 02:32:50 2025
    From Newsgroup: alt.comp.os.windows-11

    On Mon, 11/24/2025 12:50 PM, Andy Burns wrote:
    On a Win11 machine with an i5-1335U (2x P cores with hyperthreading, and 8x E cores without HT)

    For manually setting CPU affinity, is it guaranteed that CPU0 to CPU3 are the P cores and CPU4 to CPU11 are E cores?

    We can look at two different models for inspiration.

    https://www.techpowerup.com/cpu-specs/core-i5-1235u.c2591

    https://www.techpowerup.com/cpu-specs/core-i5-1335u.c3065

    Now, one of the pretty basic questions for modern CPUs,
    is do they use mesh bus or ring bus. You would be surprised
    to find they're still using ring bus, although "it is a
    bit of a loser" on performance. On my 4930K, it turns
    a 6 core CPU, into 5 cores of performance (one cores worth
    lost to the ring bus).

    This article claims you can look at /proc/cpuinfo
    for some sort of information like you are seeking.
    Yet, Linux simply does not do Task Manager displays for
    its CPU activity, so there is hardly a chance of
    transferring the info visually from one OS to another.

    https://sites.utexas.edu/jdm4372/2021/05/27/locations-of-cores-and-l3-slices-on-xeon-scalable-processors/

    You could try running SuperPI, change the affinity setting
    and see where the "lump" goes. SuperPI runs on a single
    thread, and using affinity would force it onto a single
    "core" if you want.

    If you did whole SuperPI runs by setting Affinity before
    the run starts, you could do a "bench per each core".
    You should get four results which are better
    than the other eight results. In the Affinity mask, you'd
    only enable one CPU core in each case, for the run. Then
    you could plot the SuperPI times on a representative diagram
    of Task Manager that you were watching during each Affinity
    run. That would not "number the diagram" necessarily,
    but it should help identify which are a (hyperthreaded) P.

    Paul

    --- Synchronet 3.21a-Linux NewsLink 1.2
  • From Andy Burns@usenet@andyburns.uk to alt.comp.os.windows-11 on Tue Nov 25 09:49:16 2025
    From Newsgroup: alt.comp.os.windows-11

    Paul wrote:

    This article claims you can look at /proc/cpuinfo
    for some sort of information like you are seeking.
    Yet, Linux simply does not do Task Manager displays for
    its CPU activity, so there is hardly a chance of
    transferring the info visually from one OS to another.

    Yeah, not sure Id trust what I could learn from Linux to Windows, nt
    even from Win10 to Win11.

    I managed to remote onto one of the devices, the complaint is about the
    fan noise it produces when someone sends it "some work" to do, so by
    changing affinity I can certainly get the core temperature down from
    95°C to 70°C and fan speed down from 6500 to 3700 rpm.

    The supplier claims they can't reproduce it, but I've learned they're
    testing with an unknown i7 instead of an i5, and 32GB instead of 16GB;
    no doubt the real cause is poor software but I doubt I'll impress that
    on them, so have to treat the symptoms rather than the cause ...


    --- Synchronet 3.21a-Linux NewsLink 1.2