Xeon E5 cpu work in low status
Ian Smith
smithi at nimnet.asn.au
Tue Nov 5 02:09:47 UTC 2013
On Mon, 4 Nov 2013, Adrian Chadd wrote:
> On 4 November 2013 14:24, Konstantin Belousov <kostikbel at gmail.com> wrote:
>
> > My Intel board DQ67OW starts with the fixed CPU speed, which is
> > configurable in BIOS. Unless OS starts managing the frequency with the
> > cpufreq and powerd, CPU is locked to the pre-configured speed. It was
> > not easy to understand why my single-user memory b/w benchmarks show
> > half of the expected throughput for the cache, until I found the setting
> > and found that Intel defaults to 1/2 of the marketing frequency.
> >
> > For my board, it is Performance->Processor Overrides->Maximum Non-Turbo
> > Ratio. It was set to 17, normal CPU mode is 34, turbo is 38 max.
>
> Does powerd throttle each individual core like this? I don't have
> anything laptop-y that's recent enough for that to matter.
Unless something's changed recently I've missed, no. My 9 box is still
broken, so ref to 8.2 sources .. start with /sys/kern/kern_cpu.c,
cpufreq(4) and results of find /sys/ -name "*freq*". Nate Lawson
commented way back:
/*
* Only initialize one set of sysctls for all CPUs. In the future,
* if multiple CPUs can have different settings, we can move these
* sysctls to be under every CPU instead of just the first one.
*/
I recall that some CPUs had potential for individual settings (voltage,
freq) but some only per-package rather than per-core, and then there's
the thermal control settings as well. I believe Kevin's right, p4tcc
and acpi_throttle should be disabled by default, on most CPUs anyway,
and that at least is just a simple change to hints, as a start.
It's quite a deep rabbit warren, and most docs are only in the code - eg
still? no mans for most if not all of the absolute and relative cpufreq
drivers mentioned in cpufreq(4), as I recall - and of course the moving
targets with newer CPUs .. so have fun down there!
cheers, Ian
More information about the freebsd-acpi
mailing list