Cool deal: 27" iMac 5K with 1TB Fusion / 24GB mem / Applecare only $1655
Mac Pro Westmere Real World Memory Bandwidth
The memory bandwidth test uses MemoryTester, the numbers are from the stress command. The 'stress' command uses all virtual cores at once for 100% CPU utilization. It is a pure memory-speed-bound test.
The MemoryTester number is a true measure of the ability of a real program to utilize the memory system. Small differences should be ignored as there is some fractional measurement error.
- Though 1333MHz memory should offer 25% more bandwidth to a real program, testing shows that the gain is only 10% for a real program.
- Using 1066Mhz memory in the 2010 3.33GHz Mac Pro (1333MHz is stipulated) results in identical memory speed to the 2009 3.33GHz Mac Pro.
- An 8-core or 12-core system has dual banks of memory, hence greater bandwidth.
Yes, you can use one (1) 8GB module in the hexacore Mac Pro, but the memory bandwidth is severely reduced. Not recommended, but two modules OK.
Click to view a larger graph.
Does it matter for real tasks?! Permalink
Memory bandwidth differences are hard to measure with real program doing real tasks because the on-chip caches mask the effects, and most programs are not memory-access intensive.
To find out if memory bandwidth would actually matter for common tasks, including memory intensive ones, such as Photoshop. I chose the diglloydSpeed1 benchmark, and tested with 2/3/4 modules in the 3.33GHz 6-core Mac Pro. This test has plenty of memory even with 2 modules, so the amount of memory is not a factor.
- The test took 2.7% longer with two modules as compared with three.
- The test took 2.4% longer with four modules as compared with three.
- I also tested with Helicon Focus and Photozoom Pro. No difference between 2 or 3 modules.
If you want maximum speed, stick with three modules. But remember, if memory runs low, more memory is far faster than too little and these small differences seen here are irrelevant.