From: Jerry Van Baren <gerald.vanbaren_at_smiths-aerospace.com>
Date: Thu, 05 Jan 2006 14:11:00 -0500

Couple of counter-points:
* Power consumption is current times voltage:
   P = IE
If you are using a resistor:
   E = IR, I = E/R
Therefore, _in a resistor_, power is quadratically related to current
(given a constant voltage) or voltage (given a contant current):
   P = I^2 R
   P = E^2 / R
If your piece of equipment isn't a resistor, the above may not be true
(but it tends to be quadratic). A perverse counter example is a zener
diode where the diode fixes the voltage (E) and thus the power
dissipated by the zener is linearly related to the current through the
diode.

* Power consuption is fairly linearly related to clock speed, holding I,
E, and fabrication technology the same. One reference:
   <http://www.embedded.com/story/OEG20030121S0057>
Note that Transmeta (followed by Intel and the others) control power in
their chips by scaling both clock _and_ voltage and thus their power
control is quadratic (the clock and voltage control "multiply together"
to give a quadratic result).

* Quite often clock speeds are only crudely controllable. Some hardware
give pretty fine-grained control, but most hardware only gives a couple
of choices.

* Typically software needs to run in bursts to meet deadlines and get
reasonable performance (i.e. user response time). Users tend to be very
bursty (want immediate response to key presses, but press keys very,
very slowly - relatively speaking). Users are going to be very unhappy
with a constant 2x speed when they really want
   20x 1x 1x 1x 1x 1x.... 20x 1x 1x 1x 1x 1x....

gvb


Ronald Teune wrote:
> Hmm... wasn't the power consumption quadratically related to the power
> consumption? If that's true, it would be better to have a constant
> average speed:
>
> Scenario 1: speed is 3x, 1x, 3x, 1x
> power consumption is 9x,1x,9x,1x = 20
>
> Scenario 2: speed is 2x, 2x, 2x, 2x
> power consumption is 4x,4x,4x,4x = 16
>
> Or is it impossible to boost to a certain value? Or am I making some
> calculation error?
>
> Ronald
>
> On Thu, 05 Jan 2006 12:00:25 +0100, <rockbox-request_at_cool.haxx.se> wrote:
>
>> On 1/5/06, John Lawrence <blaquestic_at_gmail.com> wrote:
>>
>>> why is the CPU boosted at times, and not boosted at other times?
>>
>>
>>> -John
>>
>>
>> Hi,
>>
>> the CPU speed is in direct correlation to the power consumption. The
>> high CPU-speed if of course needed for such "complicated" things as
>
> snip
>
>>
>> lImbus
>
>
>
Received on 2006-01-05