Quickie Performance Results for 64-Bit vs 32-Bit

Were these done with your computer booting either the 32-bit or 64-bit kernel? If you are not switching kernels, then you are testing the running of a 64-bit compiled benchmark on a 32-bit kernel. And, the Geekbench (64) app is actually a Universal Binary. Do we know if it's a true Intel 64-bit application? I'm not sure, so this issue should be addressed to PrimateLab.

If you have a computer with an EFI64 ROM you can boot into the 64-bit kernel by restarting and after the chime press and hold down the 6 and 4 keys until the spinning gear appears. You can verify if the kernel is running in the Terminal by running: uname -a.

His post shows he did 32-bit and 64-bit Snow leopard which infers he did boot up different kernels.

And I know for a fact that Geekbench 64 is intel 64..

I did not intend this as a question of your integrity. It was not at all clear exactly what you were doing. I have the most recent release of GeekBench(64) that clearly says Universal, not Intel. Hence, I wanted to know exactly what your were doing.

I have an EFI32 ROM machine and get scores of 3165 and 3466 for 32-bit and 64-bit versions of GeekBench, respectively. I cannot boot into a 64-bit kernel, yet the scores are 9.5% different. Obviously the exact conduct of such benchmarks are a factor in the results.

I think it reasonable to ask for clarification.

that is ~10% improvement between 32 and 64bit kernel... not insignificant...

About 2.5% difference between kernels:

64-bit SL/64-bit Geekbench--4,263
32-bit SL/64-bit Geekbench--4,159
(2.50% difference between 32-bit and 64-bit kernels)

64-bit SL/32-bit Geekbench--3,910
32-bit SL/32-bit Geekbench--3,809
(2.65% difference between 32-bit and 64-bit kernels)

Not much advantage to booting 64-bit kernel.

For applications the difference is larger, there is 9% difference between 32-bit Geekbench and 64-bit Geekbench.

64-bit SL/64-bit Geekbench--4,263
64-bit SL/32-bit Geekbench--3,910
(9.0% difference between 32-bit and 64-bit apps)

32-bit SL/64-bit Geekbench--4,159
32-bit SL/32-bit Geekbench--3,809
(9.2% difference between 32-bit and 64-bit apps)

The 32-bit Snow Leopard kernel can run both 32-bit and 64-bit apps. The biggest bang for the buck is running 64-bit applications.

Thanks for the geekbench test results.

Test suites however don't mean that much to me as I'm more concerned with real world performance in the apps I use the most. To that end I ran a test with Lightroom 2.4. I created a new catalog with 67 DNG’s and then timed how long it took to render 1:1 previews on all of them. I repeated each test 3 times, the results were within a few seconds of each other each time.

OS X 10.5 32 bit kernel
LR32 : 4:20s
LR64 : 3:26s

OS X 10.6 32 bit kernel
LR32 : 4:10s
LR64 : 3:08s

OS X 10.6 64 bit kernel
LR32 : 4:00s
LR64 : 3:05s

In short, 64-bit app = pretty BIG difference… 64-bit kernel = pretty small difference.

Of course gains are all depending on what your doing, large raw photo files are exactly the sort of thing where 64-bits is going to be an advantage. If you had a kernel intensive task, I suspect the OS would play a bigger role, but it doesn't here in what I spend my time doing every day.

Thank you for taking the time to do that testing. 🙂