[ltp] how to reduce the power consumption of the X200?

[Roman Haefeli]

i experience similar behaviour.

i was happily configuring and running linux-phc on my old box with a
pentium m 1.7GHz. i tested each cpu frequency seperately, while lowering
the voltage step-by-step with burnMMX (or 'yes > /dev/null' or
whatsoever) keeping the cpu busy, until the system freezed. after having
found the voltage that crashes, i used the one two steps (1 step = 16mV,
afaik) above for the config.
after having determined the lowest non-crashing voltages the way
described above, the power consumption was significantly reduced and the
cpu was running much cooler. when idling, the temperature was around
45-50°C , compared to 52-55°C from before using linux-phc. there was
even a more significant difference, when running under full load: the
maximum temperatur i was ever able to reach with linux-phc was 63°C
compared to 89° C from before. not to mention, that it wasn't necessary
for the fan anymore to run at full speed. 

so much for the success story on the old pentium m.

now, on the T61 with a Core 2 Duo 2.4GHz, things got much more complex.
my understanding of what actually is happening deep inside in the
hardware is limited, so i can only tell what i tried, eventhough it
doesn't make too much sense for someone more knowledgeable. i was trying
to use the same approach as described above for finding good voltage
values. i think, they are called 'vids' in phc lingo, where 0
corresponds to 700ms an each value grows by 16ms. i tried:
 - manually set the cpu_scaling to a fixed frequency
 - running two instances of burnMMX to keep both cores busy
 - lowering the voltage step-by-step from the highest value and wait
half a minute or so on each voltage

i could go all the way down to 0 and there was still no crash. however,
after stopping sometimes one and sometimes both instances of burnMMX, i
got an immediate crash. i thought, that i might be testing the wrong
frequency, since the fids list is a bit confusing - it looks like this:
13 12 10 8 6 136 - and it is not quite clear, which fid belongs to which
frequency. because i was unsure about that, i wrote a script, that let
me change the voltage of all frequencies simultanously. still, i
experienced the same behaviour. 

i ended up estimating good values and somehow i think it works, since at
least in idling mode, the processor is cooler than it is without
linux-phc enabled (49°C compared to 54°C). still, i think, it still
could be optimized, but i couldn't figure a way to reliably find the
optimal values.  

probably someone on the list knows how to 'correctly' determine the good
values for multi-core cpu in general or for the core 2 duo specifically.

hm.. i just noticed, that the cpu from the x200s is a low-voltage cpu
(SL9400). i wonder, if there is some room left for optimization on such
model.s


roman



On Wed, 2009-03-11 at 15:54 +0100, Fabian Henze wrote:
> Am Mittwoch 11 März 2009 12:11:52 schrieb Volker Krueger:
> > There, they reduce the power consumption by also fiddling with the CPU
> > voltage, which sounds totally crazy...
> > Can that seriously be done? What are the dangers?
> > Who has done it?
> 
> There is nothing crazy about that. As long as you just lower the voltage you 
> can't damage the hardware. However if you lower it too much your system might 
> (or will) crash.
> Sadly linux-phc is not working on my Thinkpad R61 (Core2 T7300, I can set 
> voltages to 0 0 0 0 and nothing happens, any ideas?), but I am using it on two 
> Pentium M norebooks I have here and the results are impressive :)



		
___________________________________________________________ 
Telefonate ohne weitere Kosten vom PC zum PC: http://messenger.yahoo.de