I've been sitting on some power results for a while because I haven't
had time to tidy them up but by posting them perhaps they will turn out
to be useful to someone. This email is quite long and the results are
somewhat raw and in no particular order. Rough conclusions are at the

On Linux I used powertop and on Windows I used the Lenovo battery tool
to measure draw when on battery with no mains. A "watt meter plug" was
used to measure draw through the mains with no battery physically in the
laptop. The laptop was a borrowed T60 laptop with an Intel graphics card
and a 1024x768 flat screen. The purpose of the exercise was to give me a
better idea of what draw to expect under particular conditions. I didn't
test especially rigorously but I did try and let readings settle and
where I would get multiple readings I would err on the side of least
power if I could reproduce the readings or put a range.

Unless indicated otherwise, results here are for a Gutsy Tribe 5 install
on the Linux side and an IBM/Lenovo customised Windows XP Professional
on the Windows side.

I almost certainly won't do any extra tests as I simply don't have the
time (all this has had to happen in my unpaid spare time). However
others are free to post their results and I am happy to discuss what I
did to get a particular result.

Linux power usage
(In the following results a heading will continue apply unless
overridden by a later heading)

All wake up hog programs killed, laptop mode on, backlight off
powersave governor max
16.5 wakeups per second, 11W
powersave governor ondemand
21.7 wakeups per second, 10.9W
Backlight on at maximum brightness, powersave governor ondemand
20.8 wakeups per second, 13.0W
Bluetooth and wifi on (via soft killswitch)
175.6 wakeups per second, 15.4W
Bluetooth and wifi on, compiz on
240.5 wakeups per second, 15.4W
Bluetooth and wifi off, compiz on
94.0 wakeups per second, 13.1W
Bluetooth and wifi off, compiz off (once AIGLX is turned on your
interrupts will seemingly rise until Xorg is restarted)
80.2 wakeups per second, 13.0W
(reboot, default programs running)
Brightness 100%, metacity, wifi on, bluetooth on
231.1 wakeups, 15.5W
Hardware killswitch on
69.9 wakeups, 13.2W
killall thinkpad-keys
51.7 wakeups, 13.0W
55.4 wakeups, 13.1W
kill hald-addon-storage
21.4 wakeups, 13.0W
(could try killing mixer_applet2 but it doesn't seem worth it)
101.4 wakesups, 13.1W
(blinking cursor in vi seems to use up 0.1W)
(reboot) hardware killswitch, killall thinkpad-keys, kill
hald-addon-storage, disabled laptop-mode
19.0 wakeups, 13.4W

Linux power usage on A/C
Playing music over DAAP in rhythmbox.

Windows power usage
On the Windows side this laptop came with extra tools that seem to allow
more sophisticated battery vs speed trade offs. The initial wattage
measurements are those reported by the IBM/Lenovo battery tool and are
the lowest seen reading.

Idle laptop, default battery settings
Starting notepad and typing, default battery settings

Idle laptop, optimal powersave battery settings
(this seems to allow the hard disk to still remain constantly spinning
but in perhaps in some sort of low power mode)
Idle laptop, maximum powersave battery settings
(this seems to allow the hard disk to still remain constantly spinning
but in perhaps in some sort of low power mode)
Idle laptop, hardware killswitch on, maximum powersave battery settings

(Strangely, I found that the laptop would emit a whine while it was
powersaving on battery while Bluetooth was on)

Windows power usage on A/C
The following wattage measurements were produced by my wall watt meter.

Idle laptop, wireless off, powersave maximum on A/C disk spinning
Typing in notepad, wireless off, powersave maximum on A/C disk spinning
Idle laptop, wireless on (connected to AP), powersave maximum on A/C
disk spinning
Playing music in iTunes via DAAP, wireless on (connected to AP),
powersave maximum on A/C disk spinning
Idle laptop, wireless on (disconnected from any AP), powersave maximum
on A/C
(It is worth noting that the IBM/Lenovo Windows wireless software tries
hard not to remain in scanning mode)
Playing a 3D game with sound, wireless state unknown, powersave maximum
on A/C

At least on Linux (but I'd imagine these apply to other OSes too) there
are some things you should try and do first as these seem to yield the
biggest savings. I've ordered them with the items in descending order
Disabling wifi and Bluetooth - 2.4W!
Turning off the backlight - 2.1W. If this isn't happening automatically
you might be missing out on a big saving but I've read someone saying
that laptop screens can't cope with their backlight constantly being
turned on and off so I guess caution is needed. However I've noticed
that Apple laptops running OSX seem to dim their screens fairly
Turning on laptop mode and letting the hard disk spin down - 0.4W. Again
I've been told this is dangerous as it hurts the life of the disk (wear
and tear at the most risky point) and supposedly modern hard disks have
a very good low power mode.
Using metacity instead of compiz on AIGLX - 0.1W if it's not being
masked by something else. Wakeups need to be at the 80 or below mark to
see the difference in power draw though (e.g. having wifi and Bluetooth
on will mask the effect on turning compiz on)
As mentioned elsewhere on the web using the ondemand governor may result
in less power usage than maximum throttling.

The minimum draw seen under Linux while idling in X was 10.9W (wifi and
bluetooth disabled, screen maximally dimmed, thinkpad-keys killed,
hald-addon-storage killed, mixer_applet2 killed).

Using operating system reported figures, the lowest I saw Linux go was
10.9W when idling. The lowest I saw Windows going when idling was 7.63W.
Even looking at A/C power usage via the wall plug watt monitor, Windows
seemed to use 20W while playing sound and using the wifi whereas linux
seemed to only go as low as 22W in similar conditions.

Sitsofe | http://sucs.org/~sits/

Ubuntu-devel-discuss mailing list
Modify settings or unsubscribe at: 

Reply via email to