Re: Linux 2.4.19-rc1-ac3

[Peter Osterlund]

Russell King <[e-mail]> writes:

> On Sat, Jul 13, 2002 at 09:43:16PM +0200, Peter Osterlund wrote:
> > So, is any of the above true for x86 processors? Or are there other
> > reasons to expect frequency scaling to increase battery run-time.
> 
> You're right if your CPU usage is 100% - lowering the CPU clock rate
> means you take longer to complete the task, and with the static
> element of the CPU power consumption, you'd probably end up using
> more energy to perform the same task in a longer time.
> 
> However, if, like most desktops, your CPU is sitting around 90% idle,
> if you lower the CPU clock rate, the idle time will drop.  Since the
> power drops, the rate at which the CPU uses energy also drops.
> However, overall your task completes in the same amount of time.

Hmm, assume my activity at the computer requires 10% CPU time when the
CPU is at full speed. My power consumption will then be

        P_fullspeed = P_static + 0.9 * P_idle_hi + 0.1 * P_busy_hi

If I halve the clock frequency, the computer will require 20% CPU time
to perform the work, and the power consumption becomes

        P_halfspeed = P_static + 0.8 * P_idle_lo + 0.2 * P_busy_lo

If the voltage doesn't change, 0.1 * P_busy_hi == 0.2 * P_busy_lo, so
the power savings will be

        P_fullspeed - P_halfspeed = 0.9 * P_idle_hi - 0.8 * P_idle_lo

or
        deltaP = 0.1 * P_idle_hi + 0.8 * (P_idle_hi - P_idle_lo)

The first term will be smaller the more idle my CPU is. (When reading
mail, I think 99% idle time is closer to the truth than 90%). The
second term in this formula is the reason I wondered if the power
consumption in apm idle mode is lower at lower clock frequencies.

-- 
Peter Osterlund - [e-mail]
http://w1.894.telia.com/~u89404340