Michael Pogue wrote:
> Ah, yes -- you're right.
>
> I went back and calculated the total power cost over N minutes, assuming
> a 1-minute CPU task, and in all cases, although a throttled CPU *does*
> consume less power, the task overall takes longer to complete -- a factor
> which always (for this machine) outweighs the power savings of throttling
> itself.
Which will always be the case.  Let's take a step back think about what
represents power consumption in a typical CMOS chip;  a large component is
is dynamic - pumping electrons into the gate capacitance of MOS transistors,
then draining those electrons.  Fundamentally, you start with Q = C * V to
describe what one cycle of that CMOS gate costs in terms of current, then
multiply it by the frequency of operation and you end up with I = C * V * F.
Power consumption is V * I; substituting, we get P = C * V^2 * F.  This
makes a lot of simplifying assumptions about how chips actually work.

So, T-states effectively freeze the CPU for some part of the time, thus 
reducing
the effective value of 'F' but doing no work while held.  So, a T-state 
means
that a CPU will consume (roughly) 10% less power while doing 10% less 
work in a
given period of time (there's typically some non-dynamic part of the power
consumption due to leakage, too).

P-states take advantage of the fact that CPUs running at a lower actual 
clock
rate will operate reliably at a lower supply voltage.  If you're willing to
do the work more slowly, you can reduce the V and enjoy (reduction)^2 
savings.
Just for illustration, let's assume we can crank the supply voltage down
linearly with respect to the clock rate (a drastic simplification).
Very roughly, this means that a chip might be able to consume 75% less power
while doing 50% less work in a given period of time.  As Andrei and Mike
point out, lowering the CPU clock rate and the supply voltage allow a much
greater savings in power for a given effective throughput.

So, I don't endorse T-states as a power-saving scheme; I was just observing
that a chip running at 50% clock rate will get as much work done in a given
period of time as a chip running at 100% clock rate but held 50% of the 
time.
The difference is how much power is saved.

Dana



Reply via email to