ShortShip

1 November 2008

Laptop Power Consumption

In this article we compare the power consumption of various laptops as well as look at the notion of undervolting and its affect on power consumption. Learn which features take up the most power and how to maximize your battery life on the go!


Laptops have come a long way in the last several years in terms of being a full fledged desktop replacement. However, battery life is one place that they have not made much progress in. New laptops on the market still get anywhere from 2-5 hours of battery life, which is not much better than they did 5 years ago. I recently talked to an IBM employee that stated that they were still pushing for longer battery life through lower power CPUs, battery technology and more, but I’ve heard this before. Check out this rather dated quote from a few years back.

Chipmaker Intel, in fact, is so bullish on the matter that it recently moved up a deadline for notebook makers by two years. It’s now asking them to produce by 2008 thin and light notebooks that can run for eight hours without the use of additional external batteries, said Mooly Eden, vice president of the company’s mobility group.
 - from news.com, 6/5/05

Sadly, we are really no closer to such stats, but the deadline is upon us. So, what exactly is eating up all that battery life on laptops? To figure this out, I bought myself a Kill-a-Watt from Amazon which measures power consumption of powered devices. I also went online and grabbed some similiar comparisons from other laptops (some missing information from what I measured, but overall mostly complete).

The Subjects

I decided to test my aging Acer against my new fancy Sony Vaio UX series UMPC with a solid state drive and ULV Core 2 Solo processor. While not exactly a fair comparison (15″ full laptop versus ultra portable handheld device), it should at least show the full spectrum of power consumption, old to new, big to small. I was able to find information on the web about some older Powerbooks and an ultraportible dell laptop (with a similiar processor as my Acer).

Undervolting

Another matter I wanted to test was undervolting (and underclocking). This is where you lower the voltage and/or clock speed to lower power consumption. Just like overclocking, this can result in an unstable system, so there’s a bit of trial and error. I was really curious as to what kind of power gains could be derived (in practice) from undervolting a mobile processor. Processors for laptops support running at a much lower clockspeed, which allows them, from the factory, to run at a lower voltage. The processors I tried this with supported 2 running speeds. The laptop throttles down to the lower clockspeed for normal use, and ramps up when it is needed.
Thankfully, there are programs out there that ease this process, more notably RMClock and Speedswitch XP. I decided to try RMClock.

Notes

The Kill-A-Watt is accurate only within 1W with an error margin of 0.2%. It isn’t perfect, but should be good enough. All values listed in tables are in Watts.

Results

Let’s look at some of the measurements. Note that all measurements in this table are relative to the default state (idle, full brightness, no wifi).

Model TravelMate 290 VGN-US90PS Inspiron 300M Powerbook G4 Powerbook G3
Manufacturer Acer Sony Dell Apple Apple
Screen Size 15 4.5 12 15 12
CPU Centrino 1500 Core Solo 1200 ULV Pentium M 1200 1250 G4 500 G3
Hard Drive IDE Solid State IDE IDE IDE
idle - full brightness (default) 17 10 15 23.5 15
min brightness 12.5 8 11 20 11
screen off 9 7.5 10 18 11
wifi connected 18 10 16 24.5 15.5
wifi high usage 21 13 19 - -
hard drive off 17 10 14 - -
high hard drive use 20 11 18 - -
cpu max 34 15 26 40 21
cpu max (undervolted) 28.5 14 - - -
cpu throttled 21 12.5 - - -
cpu throttled (undervolted) 20 12.5 - - -

There are several rather interesting things we can draw from the above table, the most notable being just how much the CPU draws at full tilt. I think it’s also important to note that the difference from normal hard drive to solid state, while a noticable difference, pales in comparison to the large values for CPU usage. Let’s compile all these values into per component values, for ease of comparison.

Model TravelMate 290 VGN-US90PS Inspiron 300M Powerbook G4 Powerbook G3
Sleeping 2 1 1 2 2
Charging 45 10 63 - -
full brightness 8 2.5 5 5.5 4
min brightness 3.5 0.5 1 2 0
Internal wifi (low) 1 0 1 1 0.5
Internal wifi (high) 4 3 4 - -
max usage seen 40 16 29 41 24
CPU (max) 17 5 11 16.5 6
CPU (throttled) 4 2.5 - - -
CPU (undervolted) 3 2.5 - - -
Hard drive usage 3 1 3 - -

We can see that (at an idle state), reducing the brightness of the screen can save 20%, which is actually quite significant. Converting from normal hard drive to solid state only ends up saving an extra 5-10% over conventional drives under heavy usage. I can’t see the rationalization for the extra $1000 dollars (for a sizable solid state) for an extra 5% gain. Money would be better invested in a second battery (though the solid state arguably has some performance benefit in some cases).

It is also apparent that there is a large drop in power consumption if we throttle the CPU so that it never goes into the higher (rated) speed. At high usage, the power savings are between 15-40% (with the ULV showing that it lives up to its name, as it does not appear to be as wasteful at high usage as the centrino).

However, this processor gain is only that high when the computer is at 100% cpu usage, whereas the brightness gain is all the time. Also, with running at a throttled CPU, the computer will take longer to perform high cpu calculations, so you will lose some productivity.

The following table shows the results of running SuperPI for 1 million places on both computers at both supported speeds on several supported voltages. 600mhz/1500mhz are for the centrino processor and 798/1200mhz are for the core 2 solo.

speed time watts CPU usage
600 - .956 (default) 2:12 21 4
600 - .812v 2:12 20.5 3.5
600 - .780v 2:11 20 3
1500 - 1.484v (default) 1:09 34 17
1500 - 1.308 1:10 30 13
1500 - 1.276v 1:09 28.5 12
798 - 0.938v (default) 1:14 12.5 2.5
1200 - 1.000v (default) 0:51 15 5
1200 - 0.962v 0:51 14 4

As you can see, lowering the voltage did not have any sort of effect on the speed of running the test, however lowering the voltage did affect power consumption by a fair amount. In the case of the centrino, if you were to run a processor intensive application that utilized 100% of the CPU, it would take roughly twice as long to run, but would take 30% less power, so always throttling your cpu would not make much sense if you are undervolting. If you are not undervolting, the difference is 40% less power, but 48% more processing time required, which is pretty much a wash.

What I am getting at is if the CPU needs to throttle up to a higher clockspeed, it may make your battery run longer if you prevent it, but the amount of work you will get done will be less over that time. However, most mobile use does not involve CPU heavy processes, so running at the throttled speed is probably more beneficial for more people if all they do is watch videos and browse the web.

I guess that’s about it. Feel free to post your own results or ask for any clarifications on data or results (or conclusions drawn).

No related posts.

4 Comments currently posted.

tom says:

Nice article, needs more pictures :-)

dell gx620 says:

I have heard recently Toshiba is gonna manufacture Solar laptops which would run without any battery during day time.

Udo Satiz says:

Great article that I want, however, it is a bit old at time I’m reading. Can you update for newer laptop spec?

Mohamed says:

well that was interesting, I got the same conclusion when I studied the matter :)

So its sad that there is no way to run much longer without buying another battery.

Post a comment on this entry: