OWC 16GB Memory Modules Performance on 3.33Ghz 12-Core Mac Pro (2011 variant)

2011-05-24 • SEND FEEDBACK •
Related: bandwidth, Lightroom, Mac Pro, memory, memory bandwidth, Other World Computing, Photoshop

The January, 2012 dual-rank modules run at 1333MHz, and can be expected to perform slightly better than what is shown here.

All memory modules used were OWC memory. The results shown here are with Mac OS X 10.6.7 and the latest Photoshop CS5 12.0.4 64-bit.

Memory clock speed and hence memory bandwidth is one measure of performance, but in the real world with real programs, the on-chip caches often hide memory bandwidth limitations; this varies by task because of the amount and pattern of memory access during computation.

Note: as previously documented, Photoshop CS5 is faster on a 6-core Mac Pro than on a 12-core Mac Pro, this is not new and has nothing to do with the memory, but rather with engineering assumptions that degrade performance.

Get 16GB modules here.

Photoshop CS5 12.0.4 diglloydSpeed1 benchmark

The diglloydSpeed1 benchmark runs entirely in memory on any machine with 8GB or more of memory. Photoshop CS5 should be run in 64-bit mode.

Using 6 modules is optimal for memory bandwidth; 8 modules is just a tiny bit slower in this test, consistent with past results.

Not shown is 4 X 16GB which is off the chart: 155 seconds, because the system sees 64GB as 2GB, and thrashes wildly with massive virtual memory page swapping.

Photoshop CS5 performance on diglloydSpeed1 benchmark with varying memory

Photoshop CS5 12.0.4 diglloydMedium benchmark

The diglloydMedium benchmark runs at full speed only if Photoshop CS5 has at least 18GB or more of memory available for use, because the benchmark uses ~15.7GB.

Using 6 modules is optimal for memory bandwidth; 8 modules is just a bit slower in this test, consistent with past results.

Not shown is 4 X 16GB which is off the chart: 362 seconds, because the system sees 64GB as 2GB, and thrashes wildly with massive virtual memory page swapping.

Photoshop CS5 performance on diglloydMedium benchmark with varying memory

Photoshop CS5 12.0.4 diglloydHuge benchmark

The diglloydHuge benchmark runs at full speed only if Photoshop CS5 has at least 60GB or more of memory available for use, because the benchmark uses ~56GB.

The speed of this test is driven entirely by whether there is enough memory: note that even 64GB is not quite enough.

Photoshop CS5 performance on diglloydHuge benchmark with varying memory

Lightroom 3.3 Import 128 CR2 RAW files

Import 128 CR2 RAW files, generating 1:1 high-quality previews.

Lightroom 3.3 is not demanding of the amount of memory while importing, and the bandwidth apparently has little effect on the 12-core Mac Pro, most likely because LR3 does a poor job of using all the CPU cores, so the memory bandwidth needs are relatively low, since no more than about half the cores are actually used.

Lightroom 3.3 Import 128 CR2 RAW files with high-res 1:1 previews, varying memory

MemoryTester 'compute'

The MemoryTester compute test is moderately memory intensive, and thus sensitive to memory bandwidth.

Six modules is optimal, with a clear loss of performance using eight modules. The Mac Pro uses triple channel memory (two channels of 3 modules each), so with eight modules it must drop down to dual-channel bandwidth.

Note: more than 96GB is not recognized and used.

MemoryTester compute test with varying memory