[mythtv-users] CPU power (was Possible small HD frontend)

[Brian Phillips]

Krzysztof Adamski wrote:
> There is a cpufreq daemon that does the same thing, on my P4 3.0 I
> had the frequency drop from 3000MHz to 1250MHz and I thought that I
> was saving electricity. I plugged the computer into a Kill-A-Watt
> electricity meter. At idle the box was using 91Watts, it did not
> matter if the CPU was at 3000 or 1250 same 91 W, but the box was
> slower at 1250 then 3000. Under load it would use 115 W. I did not
> measure the temperature. I gave up on the speed step.      
> 

Seems logical as the power consumption for an idle transistor is very
minimal, if completely unnoticeable.  Transistors use the bulk of power when
they need to make a state change.  The clock speed only affects the clock
signal, which is a minute number of transistors when you compare it to the
cache on chip, ISA transistors, etc.  If you clock the chip slower, you
aren't saving any noticeable amount of power because you are only reducing
the number of state changes in the clock generation and processing
transistors.  

If you are actually processing something, you are changing the state of the
majority of transistors on the chip and therefore you will see the most
power usage.  By slowing the clock, you aren't saving anything.  The total
amount of power needs to be spent to process your instructions, you are just
lengthening the time over which those instructions are processed.  You might
see lower instantaneous power usage, but summed over time you will see very
little difference in power usage.

This is all assuming minimal power lost to generation of heat, which AMD is
losing the battle on due to trying to keep up with Intel's technology.  Or
so I hear.

Brian Phillips