Revisiting "How Much Power Does My Laptop Really Use"?

And this is why I use a laptop as my local 24/7 “server” (which doesn’t do much other then keep open IRC for logs and uTorrent). You just can’t beat the power savings compared to leaving my desktop on all night.

My home desktop Core 2 Quad runs 100% constantly crunching numbers for uFluids@home. Just that in an of itself could be using over 100W!! 24 HOURS A DAY!!!

THANK YOU for this. We live off the electrical grid and use an inverter/battery set up with solar in the summer and generator charging in the winter. We bought my current computer 5+ years ago, two weeks before we found out we were moving off grid. I work from an old log cabin in the Idaho mountains, so my computer runs about 12 hours per day.

My 2.4ghz box with a 19inch CRT just EATS power, and I was trying to justify the cost of a laptop…you just did. lol - and I only started looking for info today!

(Yay! I get a new computer! lol)

Thanks,

Julie

Just some suggestions for the graph in your blog:

  • Add a legend saying what the colors are (yes, I know that information is in the text).
  • Instead of two bars showing minimum and maximum power draw, have one stacked bar.
  • If you still have the old laptop (or the power data from it), put a second series of data points showing its power draws for the same things.

Dave Lessnau

“Throttling is not always the answer”

Thats true, IF you care about whatever calculations are being done, and they’re not being done in a spin loop. If the program doesn’t yield/sleep and doesn’t terminate without user input, race to idle won’t help. I don’t want to deflate powertop’s efforts, but the timer issue is premature on the laptop. There’s other, more high profile problems facing Linux power management, like redesigning suspend and hibernate so they work every time. It’s nice that powertop suggests other measures like powering down unused USB, though.

Are the numbers at idle broken down by component? Or are the all for the same idle?

If they are the for the same idle, you could grab nircmd:

http://www.nirsoft.net/utils/nircmd2.html

and at least separate the lcd out from the rest of it, with a command like:

nircmd monitor off

This is a myth: Aero makes no difference whatsoever.
a href="http://techreport.com/discussions.x/10945"http://techreport.com/discussions.x/10945/a

The results may still apply but it’s interesting to note that they only tested desktop computers and graphics cards, so laptop hardware may be different.

With these results, it makes more sense than ever to undervolt the CPU.

http://en.wikipedia.org/wiki/Undervolting

Undervolting requires some detective trial-error work, but once done, works like a charm reducing power consumption without affecting performance.

I use Notebook Hardware Control to undervolt my Centrino Pentium-M with success, but it seems the software is not being developed anymore, and I still haven’t been able to make it work with the Core 2 Duo CPU based laptop I use at work.

“Has anyone done any testing to see if a perpetually plugged-in laptop uses power to try and charge the battery occasionally? Would it make sense for an at-home worker to remove the battery when it’s plugged in?”

Yes, when the battery drains itself to 95% and the BIOS decides it’s time to recharge to 100%, it uses power – and it shortens the battery life. If you keep the laptop on AC most of the time, you want to remove the battery most of the time, and do a refresh cycle on the battery once a month.

Those of you who leave laptops turned on 24/7 should know that standard laptop hard drives are not meant to take that kind of abuse and will die within months (Voice of experience here). Hitachi makes special 2.5" drives that are meant to be used in server applications like this, look for the ‘E’ in the part number.

An even better way to see how your laptop uses power in Linux is to run the Intel PowerTop program on it. It will show you what applications and drivers are causing CPU activity. The Fedora developers have clearly been using this tool. My laptop uses LESS THAN HALF of the power at idle than it did just 6 months ago!

By the way, suspend and resume work just great in Linux on my laptop. I had to upgrade the BIOS, though. It suspends when I close the lid, and it resumes when I open it. No muss, no fuss, works every time. It takes a while, but I have a slow disk and lots of RAM. Sometimes the USB drivers get confused after resume, but you just have to reload the kernel modules and all is fine.

If you want better suspend/resume support in Linux, complain to nVidia. They know about their bug but they have not fixed it. When they fix their bug, there will be no more excuses and everyone else’s shame will be revealed.

Somewhat related side topic: If I want to watch movies on the plane, and maximize my battery life, is there an alternative to using the DVD drive?

Can I rip a movie to disk like I can my CDs? Is any software tool that does this illegal?

How much battery life am I really saving? What’s the trade-off on disk space?

I guess a larger question, looking at Jeff’s graph:

If I rip a DVD movie to disk to save battery, how much is the CPU usage going to increase to decode the file for viewing? Does that make up for the gains not spinning up the DVD drive?

BradC, the video will be decoded by the GPU/CPU regardless of the source, computer dvd drives have no decoding capabilities, much less laptop ones.

Your best best will always be to rip the dvd to the hard drive.

I’ve found Vista on a laptop kills battery so power consumption must be higher - exact same hardware with XP gets me an hour longer. All drivers up-to-date etc…

Who knew that a solid state drive used so much power?

Any info on the difference between that and a regular 2.5" drive?

titrat: Jeff’s old laptop had a 12.1" LCD for which the difference between max and min brightness was 4W. His new one has a 13.3" LCD for which the difference between max and min brightness is 3W. I make that a factor of 13.3^2 / 12.1^2 * 4/3 ~= 1.6 in backlight efficiency.

Of course that’s pretty bogus, because (1) the measurements are only accurate to ~ 1W and (2) we don’t know how the actual brightness levels compare and (3) we don’t have a comparison to the situation where the display isn’t lit at all. So Jeff’s figures don’t exactly justify saying “see, LED backlights are better”. But if you insist on using them to try to tell whether LED backlights are better, such little evidence as there is suggests that they are.

Since someone has brought up PowerTOP and tickless Linux, I should point out http://lesswatts.org … Using powertop, laptop-mode-tools, and many of the recommendations from that website, I have managed to knock my ThinkPad T60 down to a cool ~13 watt idle.

http://www.lesswatts.org/projects/applications-power-management/race-to-idle.php

Here be hard numbers to back up the “race-to-idle is better” assertion. The basic gist of it (again, with real figures, and not my handwaving) is that the energy used by a busy, but throttled core doing some amount of work in (lets say) 10msec is greater than wide-open core doing the same amount of work in (say) 6msec and then going into a deep sleep for 4msec.

It makes sense to me, many systems run most efficiently at full-throughput; internal combustion engines generally have that property, for example. A throttled core still pays some constant power price for being on at all, a sleeping one can literally turn parts of itself completely off.

All that said, I’m sure the best scenario is some combination of dynamic scaling/throttling and sleep states. The only way to verify any optimization technique applies here as well–measure it!!

BradC, the video will be decoded by the GPU/CPU regardless of the source, computer dvd drives have no decoding capabilities, much less laptop ones.
Your best best will always be to rip the dvd to the hard drive.

A quick search tells me that in the United States, using any software that can bypass the security restrictions of a DVD is illegal, even if you are only using it for personal use.

Bummer.

Another variable to test is WiFi transmission power. Wireless cards turn down the transmission power when they receive a strong signal from the base station. Your wireless driver may allow you to override this (I know you can do this in Linux) for testing purposes.