Rosetta Performance Revisited
One of the interesting things about Apple’s transition from PowerPC processors to Intel processors is Rosetta, the technology that translates PowerPC applications so they run on Intel processors.
Whenever a system runs non-native applications using dynamic translation, there’s going to be a performance penalty. The question is, how big is that performance penalty? About a year ago, I tried to answer that question for Rosetta using a pre-release version of Geekbench 2006.
Now that Geekbench 2 supports Rosetta benchmarks out of the box (well, out of the disk image), I thought it would be interesting to see what Rosetta performance is like with Geekbench 2’s new and improved benchmarks.
So, I ran both the native and Rosetta versions of Geekbench 2 on my stock 17-inch iMac Core 2 Duo. You can take a look at the difference between native performance and Rosetta performance here on the Geekbench Browser.
If you’re not familiar with Geekbench 2, a score of 1000 is the score a Power Mac G5 1.6GHz would receive for a particular benchmark in Geekbench 2, and higher scores are better.
If you take a look at the difference between native and Rosetta performance, overall native performance is about twice as fast as Rosetta performance, even though Rosetta performance at times it doesn’t lag too far behind native performance.
What is striking is the difference in performance in the Sharpen Image and Blur Image benchmarks; these benchmarks use the same algorithms as Photoshop to apply filters to an image, so they’re an accurate predictor of Photoshop performance.
Given that the Rosetta scores are so much lower than the native scores for these benchmarks, it’s no wonder people find Photoshop CS 2 slow on Intel-based Macs. Luckily Photoshop CS 3 will be available as a universal binary at the end of March, which will fix Photoshop performance on Intel-based Macs.
So, Rosetta performance is about half the performance of native performance, which given what Rosetta is doing, is quite respectable. I know I don’t notice the difference between a universal and a Rosetta application (provided, of course, the application isn’t CPU intensive). For CPU-intensive applications (like Photoshop) you’ll certainly benefit from a native (universal) binary.