Electric energy consumption of ubuntu versus windows

[richardsdma]

starting from this topic:
http://ubuntuforums.org/showthread.php?t=2154723
i would like to hear your opinions about power consumption of a machine powered by ubuntu versus windows. this thing would only be possible if you have dual boot.
let me give you my results:
on windows ~22W
on ubuntu ~24W (only if i put the graphic card on "low" profile. if i use "default" i have 26W)
try to test using the same condition that the result to have any value.
on windows a use batteryCare and in ubuntu use powertop.

so, based on the measurements, what can we do to have the same power usage as windows?

[cariboo]

You may be seeing a difference between the different ways that Windows and Ubuntu calculate power usage, the best way to test power consumption would be to put something like this in the power line

[Bucky Ball]

You may be seeing a difference between the different ways that Windows and Ubuntu calculate power usage, the best way to test power consumption would be to put something like this in the power line

+1.

[Erik1984]

You may be seeing a difference between the different ways that Windows and Ubuntu calculate power usage, the best way to test power consumption would be to put something like this in the power line

Actually I would like to measure the energy consumption that way, I have such a meter. Too bad the wall socket is very inconveniently located (under the desk and in the corner).

[richardsdma]

it would be very nice to do such a test. i can't wait for the results.

[mastablasta]

Actually I would like to measure the energy consumption that way, I have such a meter. Too bad the wall socket is very inconveniently located (under the desk and in the corner).

use an extension cord?

[Bucky Ball]

use an extension cord?

+1. My thoughts exactly ...

[tgalati4]

I was able to get Ubuntu 9.10 to get close to Windows XP several years ago on an IBM Thinkpad T43p by doing the following:

1. Recompile the kernel to allow undervolting.
2. Set a conservative undervolt profile at all CPU step speeds (around -20% of stock voltage).
3. Set dynamic clocks in xorg.conf to allow the ATI GPU to throttle depending on load.
4. Use thinkfan to control fan speeds and allow fans to go to 0, but also go higher when needed. This results in slightly higher CPU/GPU temps (they share the heatsink) under normal use, but quieter operation.
5. Shutdown unused services.

I went from 22 watts to around 17 watts and battery life went from 2.5 hours to 3 hours. In Windows battery life was a solid 3 to 3.5 hours. So even with all of that work, I could only just meet but not exceed Windows battery life/power management. So there must be other factors like motherboard chipset power management, disk drive idle power, and perhaps wireless power savings.

You would need a special equipment to isolate and measure the power consumption of all of the subsystems.

The bottom line, you will seem some improvement, but it takes a lot of work. If you really need more battery life, buy a bigger battery (like a 9-cell) or a battery that fits into another bay.

[richardsdma]

@tgalati4
very impresive man, that is what i call a real nice job. on my machine, also with an old ati x1270, the only thing i did was to decrease the frequency of video precessor from 400 to 140 mhz using a script named powerplayswitcher, that means "low" profile. using "dynamic" profile makes the screen to flicker.

[tgalati4]

Each machine will behave differently. At -30% undervolting I was getting lockups, so there is only so low you can go before stability is compromised. I never noticed any flickering with the FireGL ATI card, but the temperature dropped 3 to 5C so that is a visible energy savings. I think the clock would switch from 400 MHz to 200 MHz. I didn't try lower, but with other utilities, you could go lower, but again at the risk of locking up the GPU--which requires a hard shutdown to recover.

My conclusion from all of this is that there are power savings to be had from all subsystems. The proprietary Windows drivers provided by the manufacturer (IBM in this case) appear to be tweaked to give the best power performance from all of the subsystems. Intel's speedstep and powertop gives a decent optimization of processes that wake the CPU, but there are a lot of subsystems that use power. Cycling the RAM clock, controlling I/O port power, North and Southbridge optimization, etc are probably needed to get to Window's power management.

I have yet to see anyone find a magic switch in Linux to improve power, so it must be difficult to achieve.

I don't know enough about how the Window's kernel operates, but I think the true multi-user, multi-tasking design of linux uses more power at idle. There is no way around those differences.