MacBook Air Mountain Lion Battery Life: 10.7.4 vs 10.8 vs 10.8.1 | Hacker News Reader

MacBook Air Mountain Lion Battery Life: 10.7.4 vs 10.8 vs 10.8.1

macobserver.com

52 pointstanousjm32 comments

I suspect something of huge impact happened in the graphics code. Maybe ML is offloading way more tasks to the GPU (it probably has to in order to optimize for Retina), maybe there are some hideous problems in the code - who knows. It's all based on anecdotal observations of course.

My Mac Pro's fan noise, while unnoticeable under Lion, has gotten so loud it's now uncomfortable to watch a movie with. It's only due to elevated temperatures in the "expansion" area where the graphics cards are (the fans on the cards are strangely quiet though). The underside of my 2011 MBP is about, say, 5-10°C hotter on ML.

That's funny, I'm noticing cooldowns across the board when using ML on my Mac Pro over Lion, although it's a Mac Pro 1,1. You might consider trying iStat Menus and using it to narrow down the problem, it gives you all the temps of the various locations in the Mac Pro. It will also let you adjust fan speed.

Make sure you reset the SMC too.

It's my fault for not being more precise. It's an early 2008 (3,1) Mac Pro with 3 Nvidia GT 120 cards. iStat is the tool I used to determine which areas/fans are running so high: it's the "Expansion Slots". Resetting the SMC was the first thing I tried, it did nothing (which is not unexpected since fan speed starts out normal after startup, then increasing to an uncomfortable 1600 RPM after about 20 minutes).

At this point I'm considering locating the temp sensor to stick a piece of gum on it or maybe disable it outright.

I'm a little ashamed to say that the other alternative is a downgrade to SL, which for the purposes of making peace with a once beloved OS I shall refer to as "The Good Old Times" from now on.

I would bet you anything it's those stupid GT120s. There is a reason you can't buy them from Apple anymore. They basically never made a decent Nvidia card. Or, wait for an ML graphics update. $20 says it will solve your problem.

The 4870/5870/6870 is a good choice if you're looking for something more compatible. The drivers for that series are steadily getting better and you can use the PC versions right out of the box with no modifications.

However if you have 3 GT120s for the reason I suspect, you need to power six displays or three dual DVI displays, then you're stuck.

Like I said it was fine on Lion and before. Fork over those 20 bucks ;)

I don't think there will be a graphics update for ML by the way. The MP (and whatever cards it uses) is an obsolete platform.

Thanks for suggesting alternative graphics, but I already upgraded from the previous gen ATI cards a few months ago and I'm not likely to buy more parts for this machine again. At this point I'm waiting what Apple does with the Pro and maybe buy a new one (or just leave Apple if they screw this product up too badly).

> Like I said it was fine on Lion and before. Fork over those 20 bucks ;)

Sorry I didn't mean there is a hardware problem with the cards, just that the hastily written ML GT120 drivers are most likely the culprit.

> I don't think there will be a graphics update for ML by the way.

Graphics updates are an expected part of an OS X release cycle at this point. We had one for SL, and Lion, and I would expect one for ML. Touchups across the board. What you're ignoring is that you already got a graphics update. By upgrading to ML, you're now on the 64-bit kernel and therefore are using the 64-bit kexts.

> The MP (and whatever cards it uses) is an obsolete platform.

Apple reaffirmed their commitment to the Pro line, so I wouldn't assume that so quickly. Also, behind the scenes evidence highly suggests that isn't the case. If Apple is planning to ditch the Mac Pro, they sure are steadily adding PC graphics card support for no explicable reason to OS X then.

Between 10.6.8 and 10.8.1 we've had support added for generic ATI cards and Nvidia cards of all stripes. Before 10.6.8, you simply couldn't boot a Mac Pro without an official card or card reflashed with EFI firmware.

Now, I haven't had an Apple official or reflashed card in my Mac Pro in a year and I use it every day, including right now. None of this has yet resulted in a product Apple actually sells. That bodes quite well for the future of the Pro line.

I'd really like to see Snow Leopard added to this, because absolutely all personal testing I've done has shown that we've been in a steady descent from that point if you care about things like battery life and performance. I simply don't have enough time to objectively measure this, unfortunately, but if you take the aggregate of various benchmarks and reviews, that seems to be the consensus anyway.

I posted a very similar reply to a post a couple of days ago linking to the same article (uh..see EDIT below, this is slightly different article), submitted by the same person, who also happens to be the the author of the story (http://news.ycombinator.com/item?id=4425673). It appears that post was either flagged or downvoted into oblivion, because it disappeared shortly thereafter (and after seeing this hit again 2 days later, I'm thinking it was pretty justifiable...)

EDIT: okay, so I missed a small detail, the other was about the MBP, this one is about the MBA. My bad..I guess.

Anyway, my experience has been largely the same as yours...I lost a significant amount of battery life when I upgraded to Lion, and apparently it seems like the problem is only getting worse.

The author replied in the other post that he would add Snow Leopard to the test, which should be interesting...I'm expecting that the difference between Snow Leopard and Mountain Lion will be huge.

Hi toadkick,

Yep, I'm currently sitting here running 10.6.8 battery life tests as we speak. I'm going to make it as comprehensive as I can and run multiple tests on each build of OS X from 10.6.8 to 10.8.2 (prerelease). It will take a while; should be ready by the middle of this coming week.

...and that is after 10.6 supposedly lowered battery life too:

http://www.anandtech.com/show/2870/done-for-2009-the-holiday...

I tried 10.5-10.8 on the same MacBook Pro and anecdotally, it really is a downward trend. Looking forward to your benchmarks.

Though 10.6 did improve overall benchmark performance from 10.5 which was arguably the first regression in OS X, which from its rough inception managed to get better by the point release. It's actually rather alarming that 10.8 is a continuation in decline rather than making up from 10.7's failings.

I switched from using OpenBSD on all my laptops at 10.3 and now I find it very difficult to continue justifying OS X with all of these performance regressions and how increasingly annoying it is to treat the OS as a solid desktop *nix, which is all I ever wanted. As much as I admire DragonflyBSD, I'm likely going to Linux next.

(What I really miss was how snappy and responsive IRIX was...)

The most frustrating thing is that both "the internet" (the blogosphere or whatever you'd call it) and sales numbers are absolutely in favor of Apple's strategy. Sometimes I need to open up the Mac App Store to look at all the grumpy one-star reviews for 10.8, just to know that humanity hasn't completely lost its mind.

Should have done a SMC and PRAM reset before performing the tests.

And of course, not forgetting repair permissions, a quick shake of the blessed ju-ju stick in an easterly direction, and the feeding of at least 2 manna cookies to the Jobs voodoo doll.

More seriously, has SMC/PRAM reset ever fixed anything for anyone?

Resetting PRAM has fixed a variety of things for me over the years, though not necessarily on newer macs.

kernel_task on my rMBP was hitting 100% CPU at seeming random times (about once a day), and it would stay there until I rebooted. SMC/PRAM reset seems to have stopped it happening.

On certain PowerBooks and the first generation Intel MBPs, yes, but they fixed those bugs with a firmware update aeons ago.

The OP (tanousjm) posted the same thing 2 days ago:

http://news.ycombinator.com/item?id=4425673

I'm relatively new to HN, but I believe the first comment on tanousjm's original submission is noteworthy (especially since tanousjm posted the same thing twice in as many days).

These are different articles, Casseres. First one was MacBook Pro battery tests, this is a follow-up with MacBook Air tests.

The way I understand Gatekeeper, it must checksum an app on disk whenever it starts up (correct me if I am wrong)

If that is true, this test may not be representative of typical use; it contains way too many application launches.

I would like to see the effect of Gatekeeper settings on battery life.

Gatekeeper only runs the first time something is downloaded and has nothing to do with it.

Codesigning does verify programs each time they are run and this has not changed in many OSes.

I don't think code was verified on launch before 10.8. I believe this because several Apple apps didn't verify (iTunes, and the Safari prerelease), and the only way I could tell was to verify at the command line. (Code with an invalid signature does give a message on 10.8.)

I checked this, but only got more confused. I made a copy of TextEdit, did all kinds of things with it (append some text to the executable, edited Info.plist and version.plist, threw away a zillion localizations, threw away the _CodeSignature directory), and never got Mountain Lion to complain. The mangled either launched or, in one case (I tried renaming the executable and adjusting the .plists to match), did not launch at all.

The only way I could get a code signing warning is by adding the com.apple.quarantine extended attribute (using xattr)

So, it looks like Mountain Lion only does a signing test for just downloaded apps, and then, I guess, only for downloads by applications that cooperate and set that flag.

I do think Apple has considered/is considering/would like to add more testing, though. Otherwise, I cannot explain the -P flag for codesign:

  -P, --pagesize pagesize
      Indicates the granularity of code signing. Pagesize must be a power of two.
      Chunks of pagesize bytes are separately signed and can thus be independently
      verified as needed.  As a special case, a pagesize of zero indicates that the
      entire code should be signed and verified as a single, possibly gigantic
      page. This option only applies to the main executable and has no effect on
      the sealing of associated data, including resources.

This is all by design, and you're surely right about Apple doing more with code signing in the future. In the past, signatures have mostly only been checked in very specific circumstances (e.g., designated requirements for two programs to be considered equivalent by Keychain and the firewall), though the architecture is fairly general, so third parties have been free to cook up their own applications. Still are, though there is no alternative to Developer ID for the Gatekeeper checks, which seems sensible from a KISS perspective, so long as reasonable workarounds exist for developers unwilling or unable to sign with a Developer ID (certainly true today).

The code-signing warning isn't a dialog, I think it just logs to Console.

It's more noticeable if the program was signed with entitlements (com.apple.private.*), some features might stop working then.

10.8.1 improved things a little bit, but running time is still short of where it was in Lion.

We only have one Mac in our test suite that can support it, but we're going to try to test the 2011 MBP with Snow Leopard and compare that to the results on the same machine for Lion and Mountain Lion.

Would also love to see a test of 10.7.0 as a number of us have already been experiencing battery problems since switching from Snow Leopard to Lion that were (partially) sorted out with 10.7.4 (See: https://discussions.apple.com/thread/3194235)

How many tests do you run with each configuration? I'm wondering what sort of variation these test have, but it doesn't appear to be discussed.

You're right; I'm sorry I missed discussing that. Our testing methodology is to run each test at least 2 times and if the second test is within 5 percent of the first, to average the two results. If there's a greater than 5 percent difference, we run multiple tests until we can determine which result was abnormal and why.

It's not THE most scientific methodology, but it's a decent balance between time and accuracy. In the case of our most recent round of tests, none of the second results were more than 4 percent off the first test.

I'm really pleased to hear that. Given the lack of discussion I had assumed the worst, but I'm glad to be wrong, and sorry to presume otherwise. Thanks a lot for following up.

I hope this makes some kind of a difference on my 2010 Air. 10.8 really destroyed the battery on this thing.

10.8.1 seems to have improved my battery life on my MBA from 10.8. 10.8 really killed the battery life. But 10.8.1 I'm noticing improvement.

Reply on news.ycombinator.com