[CentOS] Intel turbo mode

Mon Oct 9 12:58:55 UTC 2017
hw <hw at adminart.net>

Stephen John Smoogen <smooge at gmail.com> writes:

> On 3 October 2017 at 13:01, hw <hw at adminart.net> wrote:
>> Stephen John Smoogen <smooge at gmail.com> writes:
>>
>>> On 1 October 2017 at 11:34, hw <hw at adminart.net> wrote:
>>>> Hi,
>>>>
>>>> is there a way in Centos to find out if the Intel turbo mode will be
>>>> used?
>>>>
>>>> Using the 'stress' utility and checking the frequency with cpupower
>>>> tells me that a CPU is running at it´s maximum frequency as reported by
>>>> cpupower --- and this frequency is less than the frequency it would run
>>>> at if it used the turbo mode.  All the other CPUs are at their minimum
>>>> frequency.  I have verified that turbo mode is enabled in the BIOS.
>>>>
>>>> Is cpupower unable to report frequencies used in turbo mode despite it
>>>> always says it gets its information from the hardware?
>>>>
>>>
>>> It would seem that there are multiple ways to get the information you
>>> are looking for. I expect you have seen this already
>>>
>>> https://haypo.github.io/intel-cpus.html
>>>
>>> but I figured I would pass it on for others. They found that cpupower
>>> is less reliable in how it reports the data because of the values it
>>> gets them from.
>>
>> Thanks, I didn´t see that one yet.  At least I noticed that cpupower
>> says that turbo mode is available, and turbostat seems to indicate that
>> CPUs sometimes run at higher frequencies like they would when in turbo
>> mode.
>>
>> It´s strange that there is no tool to definitely figure this out,
>> especially since RH seems to have done a lot of research into improving
>> performance.
>>
>
> My limited understanding is that it isn't very reliable to show it and
> the Windows ones distort reality a lot (aka say you are in it when you
> aren't actually in it) because they do a moving average to show what
> is going on so it doesn't look as jagged as it really is.

I guess the same could be said for turbostat because it show some
computed frequency.

> It is also not all that useful for general software needs. The CPU is
> going to be waiting a lot longer now for memory and io to catch up for
> most transactions.

Does it hurt anything when it waits faster?  It might have the advantage
that it doesn´t wait longer than it otherwise would and processes things
somewhat faster when it finally does.  For how long it takes until
processing is finished, it doesn´t matter if the CPU waits longer when
it still takes the same amount of time, and chances are it might not
take as long.

Or am I mistaken?

Anyway, it seems to me as if there´s actually no way to figure out if
the CPU is at turbo frequencies because it decides that for itself.  But
then, I have an E3-1230 V2 here, and even cpupower reports its turbo
frequencies.


-- 
"Didn't work" is an error.