The dGPU: Killing Battery Life

The 15 and 17-inch MacBook Pros have a discrete GPU that only turns on if you fire up an application that really needs it—at least that's how it is supposed to work. In practice, the discrete GPU takes over control if your application uses any one of a number of frameworks—and some of the time, the dGPU simply isn't necessary.

Case in point, launching Chrome won't trigger a dGPU switch but the moment it encounters Flash the discrete GPU will take over. The bad news is that even if you close all Chrome windows, the dGPU won't power down until you quit chrome entirely. The same is true for Photoshop. Launch the application and you're still on the iGPU. Actually open up an image and the dGPU takes over. Even if you close all open images and just leave the Photoshop application open, the dGPU won't relinquish control. FaceTime and anything using the integrated camera also require the dGPU, despite it being totally unnecessary.

If you connect any external display to the 15 or 17-inch MacBook Pro that also forces the dGPU on, at which point both the integrated panel and external display are driven by the dGPU. There is no funny frame buffer copying going on, both the integrated and discrete GPUs have their own connection to the display.

Apple also fails to provide a way of turning off the dGPU by default—the best you can do is shut off the iGPU and just use the dGPU entirely. Thankfully Cody Krieger's gfxCardStatus tool gives us exactly what OS X does not. Version 2.0.1 adds support for the 2011 MacBook Pros.

I'm going on and on about the dGPU because it's state can seriously impact battery life. The numbers below should help put that in perspective for you:

Impact of Discrete GPU on Battery Life
15-inch 2011 MacBook Pro Light Web Browsing Flash Web Browsing
Integrated GPU (Intel HD 3000) 8.85 hours 7.03 hours
Discrete GPU (AMD Radeon HD 6750M) 5.67 hours 2.97 hours

Even just browsing the web, the dGPU being on drops battery life by 35—60%. Under full CPU load I suspect the percentage difference would be smaller, but still significant. The worst part of this all is that without gfxCardStatus you can negatively impact battery life by doing something completely innocent like accidentally leaving an application open. Given how much OS X is tailored to simply closing windows when you're done with them and not quitting applications, an overly aggressive dGPU can really be an issue.

Thankfully we do have gfxCardStatus but there's honestly no reason Apple shouldn't include this functionality with OS X from the start.

The GPU Comparison Display Quality
Comments Locked

198 Comments

View All Comments

  • zhill - Friday, March 11, 2011 - link

    Hmm... so I mostly disagree, but the fact that there is some confusion is problematic as well.

    I assumed (the need to assume is the problematic part) that the main battery life tests (Web loads, etc) were run with OSX doing whatever it wants. The point of the gfxCardStatus tests was to specifically point out the difference in power consumption with each card being used. The only way to expose that behavior explicitly is to manually enable/disable the dGPU.

    So, I think the methodology makes sense, but I agree that Anand should make it clear in the general battery-life section that OSX is managing the GPU in stock form with no gfxCardStatus inferference.
  • 7Enigma - Friday, March 11, 2011 - link

    I ran the numbers. The values (in minutes) that were reported in the 1st and 3rd charts of page 15 which correspond to the light and Flash-based web surfing are dead on the bottom of page 9 numbers where the discrete GPU has been TURNED OFF.

    It is so bad that after running the numbers if you were to use the data from page 9 the 15" MacBook Pro would be so far in last place on the Flash-based chart (at 177minutes) that the next highest is over an hour and a half LONGER.
  • zhill - Friday, March 11, 2011 - link

    I believe he's using Safari in the web and flash web tests, and mentions:
    "Another contributing factor is the new 32nm iGPU which is active full-time under Safari. " When discussing the Flash problem (on pg 9) he specifically mentions Chrome + Flash activating the dGPU.

    So, the question is: what is the dGPU behavior for other browsers? Do Firefox and Safari only use the iGPU or is the dGPU activation only in Chrome? It needs to be addressed in the article more fully for no other reason than clarity and so users will know that choice of browser may impact battery life very significantly.
  • 7Enigma - Friday, March 11, 2011 - link

    That is a good point (missed the Chrome/Safari switch), but it is also puzzling that the numbers from both charts match up perfectly. I would have expected a bit of difference between Chrome and Safari even if the dGPU issue is taken away just due to coding differences, but if you divide the minutes by 60 from the earlier page you'll get the EXACT same number of hours posted in the later battery numbers. The only way to describe this without it being the same number would be if they are comparing numbers from 2 different browsers in the same chart without labeling as such which I find very hard to believe.
  • 7Enigma - Saturday, March 12, 2011 - link

    Seriously not a single reply from one of the authors?
  • IlllI - Friday, March 11, 2011 - link

    theres hardly much difference between the 13in model and the 17in. so basically you are paying $1000 more for 4 inches.
  • alent1234 - Friday, March 11, 2011 - link

    i5 to i7, discrete GPU, more hard drive space

    almost same price difference if you went with dell/hp
  • khimera2000 - Friday, March 11, 2011 - link

    Just so you know looking at the recommended configuration of hp envy cost 1932, a maxed out 3D vision dell xps 17 cost about 2480 but has a sell that brings it down to 2244 (dual HDD no SDD) where as the base configuration of the mac cost 2500. so..,. no even if you throw in the discreet, I7 increased harddrive space your still over the mark when you compare to windows notebooks.

    How did i find these numbers? opened up the web sight for each respective company and looked at what they had as of 5:00 PM 3/11/2011 applicable to US customers (after all they where the US versions of the sight) except for dell no others where offering an automatic discounts.
  • jed22281 - Friday, March 11, 2011 - link

    Should've done the 13" and then jumped up to the 17".
    Is there some reason you're not interested in the 17" versus the 15"?

    Thanks!
  • jed22281 - Saturday, March 12, 2011 - link

    Anand, Brian, or Vivek?

    Thanks!

Log in

Don't have an account? Sign up now