[CentOS] Questions on cpu frequency scaling AMD vs. Intel

Mon Aug 4 13:51:41 UTC 2008
Ned Slider <ned at unixmail.co.uk>

Kai Schaetzl wrote:
> Kai Schaetzl wrote on Sun, 03 Aug 2008 16:57:20 +0200:
>> I disagree about the reason. I think they are actually not so efficient. At 
>> least not if I compare to a low-voltage CPU.
> Just checked how much that AMD 4850e CPU drains under various conditions. 
> There are *huge* differences. I checked whole power consumption of the 
> machine. I don't know what "at wall" means. Did you measure the power 
> consumption of the cpu alone or does "at wall" mean the same as I did?

Yes, sounds like you did the same as I did. I meant I plugged one of 
those watt meters into the power outlet at the wall and plugged the 
machine into that, so you're measuring the current draw "at the wall" or 
outlet. What this doesn't do is take into account how efficient (or 
inefficient) your power supply may be - if it's drawing 100W from the 
wall and is 80% efficient, then your system is only actually pulling 
80W, the other 20W is heat dissipated from the PSU.

> Here are the figures, considering this is for the whole machine I think it's 
> quite good.
> idle:
> 1000 MHz: 76W
> 2500 MHz: 98W

That's a nice little saving! Like I said previously, I only saw 2-3W 
saving at idle between full clock rate(2400MHz; 107-8W) and with freq 
scaling active (1600MHz; 105W) which would maybe imply that my system 
already has efficient halt state, and that throttling back (freq 
scaling) gives little further gains. Obviously that's not the case with 
your system.

Were you able to observe any drops in VCore voltage between load, idle 
(2500MHz) and 1000MHz with lm_sensors?

> 1 core under load: 110W
> 2 core under load: 120W
> So, that's not just the processor, it's the whole machine. It takes into 
> account the powerdrain from the processor plus (probably) faster fans plus any 
> other drain from memory/chipset that may be higher underload.

Likewise :)