Wakeup/s or avg C state residency

[Ganesha Bhaskara]

Arjan van de Ven wrote:
> Alan Jenkins wrote:
>>
>> AFAIK: The headline figure would be %age time in C4. 
> 
> yup
> 
>> I'm thinking the avg residency figure doesn't actually add anything to
>> that. 
> 
> that's not quite the case unfortunately.
> 
>> Transitioning to/from C4 takes some time, during which power is being
>> drawn but no productive work done.  However if you spend more time in
>> transitions but the same time in C4, all that means is you got less
>> time doing productive work; there might be no difference at all in
>> terms of power usage.
> 
> transitions are actually quite expensive (well everything is relative of
> course); it means analog parts of your chip need to go on or off. Think
> of it as with traditional light bulbs; when they turn on they briefly
> take a lot more current than when they're fully enabled.
> (this is also why they break at this point more often)
> 
> Depending on the cpu type, the time you want to be in C4 ranges from
> 20msec to 50msec. (this assumes current cpus; future cpus might and will
> vary). Above 50msec you'll not see much power difference today.
> 
> 
> So the answer to the original question is something like:
> * try to get C4 %age above 99%
> * try go get average time there above 50msec

Thanks ..... exactly what I was looking for. I guess this would be a
good data point that could be added to the FAQ

-G