Processor Scaling

[Derek Broughton]

Rashkae wrote:

> Derek Broughton wrote:
> 
>> 
>> Load.  You're running it at its maximum _frequency_, but the processor
>> itself will be idle some percentage of the time.  aiui, a processor
>> running at 50% load and 100% frequency, is using more power than one at
>> 100% load and 50% frequency.
> 
> You got data to back that up Derek? 

Not a bit - as I said, that's my understanding... but I was wrong once. :-)

> When I tested my PC (Core 2 Duo)
> with a power meter, there was only 5W difference when I changed my CPU
> frequency from scaled to full frequency when the cpu was idle.  However,
> putting the cpu under load would spike the power by over 15W.  (I did
> not, however, try to force the CPU at full load / half frequency)

Yeah, that's not telling you (or me) an awful lot.

You'd need to measure both frequency and load while doing the same tasks - 
and you probably need to try to use a task that tasks the CPU but not to the 
limit (because if it's using 80% cpu at full frequency, and 100% at 50% 
frequency, you don't know just how high it _could_ have gone).  And you'd 
also want to know if the task took longer at the lower frequency.