OWC 16GB memory modules for 2009/2010/2011 Mac Pro
In January 2012, OWC released the 2nd generation of their 16GB memory modules, which now run run at full-bus speed of 1333 MHz*, and perform as well as 8GB memory modules.
According to OWC—
These 16GB modules allow full 1333Mhz bus performance and are faster in both 1066MHz and 1333Mhz memory bus Mac Pros with the Dual-Rank design than the Quad-Rank module options they replace.
I tested the OWC 16B memory modules on both a 6-core 3.33 GHz and 12-core 3.33 GHz Mac Pro. The modules on both machines showed up as 1333 MHz, and tested at the same speed as the 8GB memory modules, using MemoryTester with the stress command, which display memcpy() memory bandwidth, a real-world test.
* OWC released the first variant of their 16GB memory modules for the 2009/2010/2011 Mac Pro in May, 2011. These modules worked great, but did have one downside of downclocking to 1066 MHz or even 800 MHz.
A 4-core or 6-core Mac Pro can accept up to 48GB.
An 8-core or 12-core up to 96GB.
More than that is recognized in hardware, but Mac OS X Lion 10.7.2 will not use more than 48/96GB respectively. OWC tells me that the same Mac booted into Windows boot can address 128GB, so this is clearly a Mac OS X bug, not a hardware limitation.
Using 64-bit MemoryTester, the alloc command can allocate about 91GB of that 96GB before severe virtual memory paging begins. So the absolute limit to addressable memory for a single program about 91GB, but it would be wise to limit any program to 85GB or so.
Cut those figures in half for a 4/6-core Mac Pro which can take 3 X 16GB memory modules: about 43-44GB of 48GB, with a realistic limit of 40GB.
Mix and match
Current iterations of the 16GB module don’t “play” with the 8GB modules— ask OWC if unsure.
Optimal setup for maximum memory bandwidth remains triple-channel memory configuration:
- 3 modules in the 4/6-core Mac Pro (fill 3 of 4 slots), 48GB max.
- 6 modules in the 8/12-core Mac Pro (fill 6 of 8 slots), 96GB max.
However, the real-world impact of triple channel vs dual channel memory bandwidth is very small, so this can be safely ignored for practical purposes, and if the additional memory is actually needed, then more is far faster.