Snow Leopard performance improvements are there, but small

Snow Leopard is purported to provide many small but much-needed tweaks to its predecessor, Leopard. One oft-touted tweak is a speed boost, but according to tests by Macworld the performance and speed of a few different computers improved only slightly with many native tasks, and some took even longer.
Macworld installed Leopard and Snow Leopard on even-sized partitions on the drives of three different configurations: a 20-inch 2.66GHz iMac Core 2 Duo with 2GB of RAM, a 3GHz Xeon 5300 eight-core Mac Pro with 4GB of RAM from April 2007, and a 15-inch 2.8GHz MacBook Pro Core 2 Duo with 4GB of RAM. They charted simple tasks like start up, shut down, PDF scrolling, and more complicated ones like iMovie import/export and Photoshop CS4 filters. You can see the final results here.
The chart is a bit confusing about the actual speed improvement, and it is important to note that a mark of 100% on the chart indicates that the task performed was the same on both operating systems; likewise, a mark of 103% means it the task was 3% faster with Snow Leopard, and so on.
The improvements were small on most fronts, and the only significantly improved tasks were shut down, JavaScript, and Time Machine. The MacBook Pro with Snow Leopard inexplicably saw a huge improvement of 42% over Leopard when it imported movies into iMovie, while the the other two computers barely budged. A few of the benchmarks were even slower with Snow Leopard, such as waking the computer up and opening duplicate Finder windows. While the tested computers only represent a small part of the spectrum, it appears that now Leopard's speed improvements for native applications are there, but not mind-blowing.

![TUAW [Cafepress]](http://www.blogsmithmedia.com/www.tuaw.com/media/tuaw-cafepress-promo.png)


Reader Comments (Page 1 of 2)
EMoShunz said 1:40PM on 8-27-2009
was that snow-leopard system an upgrade, or clean install. they should maybe compare both... at least i'd like to see that.
Reply
Hunter said 1:53PM on 8-27-2009
In either case I doubt they waited for the mdworker background process to catch up. It consumes a fair amount of resources after install to build meta data in the system.
Reply
John said 1:57PM on 8-27-2009
Regarding installation of Snow Leopard-I was just listening to Macworld Podcast and some of the editors were discussing their experiences with SL. One of the things they mentioned was that you can install it WITHOUT Leopard being on your computer and that if you should want to go from Tiger to SL that you SHOULD buy the full box but you DID NOT have to. They said it would install fine over Tiger. This is the first I have heard of this.
Reply
Steffo said 1:55PM on 8-27-2009
Those results are totally flawed. The installed on the same drive which had been partitioned, which means that the first partition (presumably Leopard) would have better performance than the second partition (Snow) when it comes to disk access. So start up times and file copy times were going to be with the second OS at a handicap.
They should have done all the tests in Leopard, then wiped the drive and done them again with Snow.
Also, its quite likely that they weren't even running Snow in full 64-bit mode given the recent reports that Snow will often default to 32-bit for no reason.
So while those tests are nice and all, and show that Snow is faster on the whole, they're still fairly flawed.
Reply
Joanna D said 2:43PM on 8-27-2009
We've had random access disks for years... and years!
It doesn't matter what partition things are on - they will both run at the same speed.
Steffo said 3:02PM on 8-27-2009
Yeah, Random Access, but that doesn't mean that all access is going to be equally fast.
Simply the outer edge of the disc is moving faster. That's why when they do a test of a HDD they test the HDD's speed as it gets fuller and fuller. see here:
http://macperformanceguide.com/Storage-Drive-WesternDigitalScorpioBlack320GB.html
If you go to "Speed across the drive (area test)" and look at it you'll see that at 50% (the above review says the partitions were of equal size) it's already lost 10MB/s write and 13MB/s read speed.
I hope I've managed to give you enough evidence to show that the tests they performed were flawed. They are about 1/8th slower than they should have been. Not much, but enough to pull even or ahead in the 1GB file duplicate test and wake from sleep, and be faster at start up about 10%.
Which would effectively remove all the black spots from the Snow Leopard's record (not fur, mind)
Poltras said 3:05PM on 8-27-2009
Joanna, ma'am, you need to learn about differences in speed between the center and the edge of a HDD. Those make a large difference (seriously), actually, and it's even more important to consider when benchmarking operating systems. Random access has nothing to do with this mechanical flaw.
I'm not totally sure they are "totally flawed" as steffo pointed, but this is something to think about.
Joanna D said 3:44PM on 8-27-2009
I wasn't aware you could control the physical location of ANY of the data on your disk - so why do you think that the partitions will always be arranged in the same pattern?
punkassjim said 4:04PM on 8-27-2009
You can't, but each partition must be comprised of contiguous blocks. Partitions don't inter-mingle with one another, so these other folks do seem to have a point. The only way to do a "clean" test would be to pit two identical systems against each other with clean installs, and enough time for post-install processes and indexing to complete.
Steffo said 4:07PM on 8-27-2009
That's just how the partitioning programs work. The first partition is on the outer edge, and then subsequent partitions move inwards from there.
Just in terms of maintaining the integrity of the partitions you can't have two partitions mixing with one another, especially if they are formatted differently and your OS can't read or can't write to one of those partitions.
Within those partitions, though, you can't control the position of the data, but generally that will be written sequentially so that when reading back the heads don't have to move much, and everything will be quicker.
This is why I always do a erase and install. It streamlines the OS's boot process.
Poltras was right, I did over exaggerate the flawed results in the tests, but any disk access based tests will be quite significantly slower as a result of this mix up.
Max Howell said 2:00PM on 8-27-2009
Call me when these shitty magazines employ some real scientists to do their tests.
Reply
David Cohen said 2:25PM on 8-27-2009
Your assessment of the numbers emphasises the 2-3% gains - while there are some pretty substantial high percentage point gains for all of the machines on some of the tests, that you fail to mention.
The one you do outline, the MBP iMovie import, you talk as being inexplicable - you have read the Snow Leopard improvement documentation, haven't you? About the ability to use the graphics processor to speed up number crunching? As the most recent machine to be launched in the test set, it stands to reason that a MBP would probably be best set to use its NVidia chipset to speed up the number crunching associated with iMovie conversion, so it is actually fairly explicable.
This sort of post is a bit premature, IMHO. Let's get the thing installed on more than three computers before we start trashing Apple's marketing, shall we?
Reply
McStever said 2:19PM on 8-27-2009
I understand that people are seeing a 10 to 15 improvement when Snow Leopard is running in 64bit mode over 32bit mode.
My guess would be the MacBook Pro was the only one with the right GPU hardware to assist with the movie import.
Reply
Jax said 2:31PM on 8-27-2009
I wonder what the margin of error was on these tests? And what was the variance? Did every finder duplicate recorded take the same length of time for example? I think some of the places where SL looks like it's slower might be within the margin of error.
I agree with the guy who said they need some real scientists doing the testing.
Reply
Joanna D said 2:40PM on 8-27-2009
Of course, because if someone says something negative about Apple then they MUST be wrong!
Jax said 12:51PM on 8-28-2009
I didn't say that, but as a scientist, I would like to know more about how these tests were performed before I trust the results, regardless of what the results show. Is there anything wrong with that?
Sam said 2:57PM on 8-27-2009
MBP iMovie import improvement could be down to using the graphics card more for stuff that requires the CPU in (not snow) Leopard
Reply
Joanna D said 2:40PM on 8-27-2009
I cannot wait to come back here on Friday and read all of the comments complaining about how Snow Leopard literally changes nothing for people, except for creating more software crashes and making the context menus in the dock black instead of white.
$29 well spent.
Reply
Jordan said 4:11PM on 8-27-2009
I'm waiting for it too. I've got it ordered, but I won't be able to install it until at least Sunday so I'll have some time to see what hell others are going through.
punkassjim said 4:17PM on 8-27-2009
I'm sure you're correct: people will completely fail to understand what they've paid their $29 for. They'll completely forget that Apple lowered the price by $100 for the express reason that there really aren't many new things in there that are readily apparent or compelling.
But that's not the point of Snow Leopard. An operating system re-write of this magnitude is a "plan for the future" type of move. Not only that, but Apple has been trying to get developers to adopt Cocoa for years, even though Apple themselves hadn't ported the majority of their code off of Carbon. Probably because of this, there are some things that COULDN'T be done with Cocoa frameworks, only in Carbon. That needs to be fixed, and I'd imagine Snow Leopard is a catalyst for forward movement.
Me? I'd gladly pay $29 simply to re-claim the gigabytes that all the Rosetta and PPC compatibility crap used to suck up. (though, full disclosure, I'm eligible for the up-to-date program, so I'll only be paying the $9.95 fee)