[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: [PATCH] powerdevil-remove-cpufreq.patch
From:       Andreas Hartmetz <ahartmetz () gmail ! com>
Date:       2010-03-08 19:48:43
Message-ID: 201003082048.43810.ahartmetz () gmail ! com
[Download RAW message or body]

On Monday 08 March 2010 20:10:16 Sebastian Kügler wrote:
> On Monday 08 March 2010 17:15:18 Martin Sandsmark wrote:
> > On Monday 8. March 2010 13.39.16 Sebastian Kügler wrote:
> > > If the ondemand governor doesn't scale up when more CPU cycles would be
> > > needed, that's a kernel bug and shouldn't be papered over by UI options
> > > in PowerDevil. For the downscaling, that is bogus anyway since doing
> > > the work as quickly as possible and going back to sleep to save power
> > > is more efficient than doing the work slower and thus keeping the CPU
> > > awake longer.
> > 
> > You're ignoring the fact that increasing CPU frequency also increases the
> > heat  generated, and at least here it then leads to the fan firing up,
> > bringing down battery time rather drastically.
> 
The average Laptop cooling fan draws about 1 watt, that's usually not very 
relevant.
Fan noise can be relevant.

> Not sure how much your fan eats, but it's a rule of thumb that doing a task
> on the CPU very fast and then sleeping for as long as possible is way more
> efficient than spreading the work over a longer period of time, working a
> little less hard.
> 
> Basically, the difference between a sleeping CPU and an active CPU is huge,
> while the difference between a fully active CPU and a throttled one is
> nearly linear (not an order of magnitude different, in any event).
> 
> Thermal management is a valid reason, though in most cases cleaning out the
> system fan will give you both, longer battery life *and* a more silent
> system. :)

The Celeron-M in my old R50e Thinkpad will be throttled, i.e. intermittently 
switched off, when "clocked lower". That does not reduce the power/cycle 
indeed.
The more expensive mobile CPUs, however, run with reduced core voltage when 
clocked lower, *actually reducing* power/CPU cycle. Above some threshold power 
consumption is roughly proportional to the square of clock frequency because 
the core voltage has to be raised to keep the CPU stable, and there are more 
cycles per second. So power per cycle increases roughly linearly. While core 
voltage does not increase much in absolute terms, what counts for power 
consumption is (core voltage - voltage drop of silicon diode). The voltage 
drop is roughly 0.7 volts.
For example, there are ultra-low voltage Pentium M / Core / whatever CPUs from 
Intel that only draw a few watts (single digit) and run at voltages around 1V. 
At least there used to be last time I looked.
Now of course the rest of the system also draws power and it *is* often 
beneficial to do something as fast as possible and then sleep. But it's not as 
simple as "reducing clock speed is always wrong".

That said, it is wrong most of the time, so we should ensure that users that 
don't know any of the above will do the right thing.
"Keep cooling fan off" (reducing clock speed as necessary) could be an 
appropriate configuration setting, for example. And rename "Xtreme powersave" 
(sic) to "Throttled", and then explain in a sentence why that isn't always 
good.
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<

[prev in list] [next in list] [prev in thread] [next in thread]