I don’t have an Intel powered Mac here yet; the old family finances just wouldn’t look very favorably down on buying one, despite my temptation and sometimes my impatience. As you might guess, I’ve been following along with the MacIntel saga as much as I can and have been especially interested in the various performance benchmarks that have been surfacing around the Net. If you’re interested in Macs and especially interested in Mac performance, there’s hardly a better place on the Web to go than Barefeats.com.
One of the great things about the website is that various performance tests are performed and results posted and left up for sometimes years at a stretch. One of the questions I had was how the new Intel Mac performance stacked up against the bevy of G5 Macs released over the years. Unfortunately, there isn’t one compilation of that kind of data at Barefeats. So, I decided to see what I could do to perform one. I only have one compilation done right now, and it centers around a test script Barefeats runs using Photoshop CS2 Actions. I compiled this data using multiple graphics currently posted on Barefeats and checked it out with Rob-art, the proprietor of the website, who ran the tests. There are several assumptions I made while performing the compilations. They are:
(1) To keep application performance the same, I used data only compiled with Photoshop CS2. Photoshop CS benchmarks were not used. I also assumed that any benchmark using Photoshop CS2 used the same scripted actions. (It turned out that wasn’t quite true, but Rob-art sent me the corrected times to account for the differences.)
(2) I assumed that performance differences due to operating system versions were minimal.
(3) If I found multiple benchmarks for a certain model Mac, I used the fastest one available.
Here’s what I came up with after incorporating the corrections Rob-art sent me:
There are several things I find interesting about these results.
First, if megahertz didn’t matter, the how come the results got better as megahertz increased? Of course, the industry is now moving away from that (Jobs') denial and, ironically, proving it true as they substitute multiple cores for increased clock speed.
Secondly, if I assume that Photoshop performance in XP is approximately the same as native Intel application performance will be under OS X, then the Rosetta speed penalty is about 67% for Photoshop. If you’ve read this blog before, you’ll remember that I was worried about Photoshop’s performance under Rosetta and did not believe Job’s claims. These numbers bear out what I was concerned about. That said, one can argue it’s still not that bad, as Jobs has. Performance using a 2.0 GHz Core Duo Intel iMac or MacBook Pro seems roughly equivalent to the current day performance of my 1.5 GHz G4 PowerBook going by these charts. Which has to make one think about Apple’s previous performance claims in hindsight…
Thirdly, continuing with the Windows XP/ Mac OS X equivalency assumption, once Intel native versions of Photoshop show up (as in Adobe Creative Suite CS3 which will be a Universal Binary), the current MacBook Pro will be roughly as fast as today’s Quad 2.5 GHz G5 PowerMac, at least in some applications. Indeed, this “order of magnitude” speed boost seems validated by the performance benchmarks surfacing at another Mac related website, i.e., Creative Mac. They’re using the slightly faster 2.16 GHz Core Duo MacBook Pro there but tests with Final Cut Pro rendering are showing it almost as fast as the Quad. So do these tests using Photoshop CS2. Other tests using Cinebench and Lightwave didn't show that kind of performance, but still demonstrated an equivalency with a dual processor 2.0 GHz G5. What does this say for the upcoming PowerMac releases? They’re going to wicked fast compared to the machines we have now. I suspect, too, they will be the Mac aficionado’s door into true 64 bit computing.