'08 Mac Pro Memory
Where you put it is just as important as how much you have.

For quite some time now, the rule of thumb for RAM has pretty much been “buy as much as you can afford.” For the most part, this axiom still rings true - the more RAM you put in, the more apps you can run without hitting speed-sapping Virtual Memory, and there’s always the many graphics and video applications that will make use of as much RAM as you can throw at it.

With the new 8-core Mac Pros, however, a new angle is added to the story - while amount of RAM is important, the speed of the RAM’s throughput is determined mostly by its arrangement within the riser cards.

The Test & Results

We tried sixteen different RAM amounts and configurations and tests were performed with a special 64-bit parallel multi-threaded version of the STREAM benchmark test created by the author of Geekbench to squeeze out the maximum throughput of the memory bus as provided to us by Rob-ART from Bare Feats. This utility measures the data throughput of the memory in this configuration.

We ran these tests on a Quad-core 2.8GHz Mac Pro as well as an 8-Core Mac Pro. Both systems were running Mac OS X 10.5.1 of a Seagate Barracuda 7200.10 SATA hard drive.

Each test was run 3 times and the score averaged to get the results here.

	Test 1	2.0GB - Stock 1GB x 2 as factory installed (1 in each riser)
	Test 2	2.0GB - Stock 1GB x 2 - but move to a single riser as pair set.
	Test 3	4.0GB - 1GB x 2 set in each riser
	Test 4	6.0GB - 1GB x 2 by 2 in top riser, 1GB x 2 in lower riser
	Test 5	8.0GB 1GB x 2 by 2 in both risers
	Test 6	4.0GB - 2GB x 2 in top riser.
	Test 7	8.0GB - 2GB x 2 in each riser
	Test 8	12.0GB - 2GB x 2 by 2 in top riser, 2GB x 2 in lower riser
	Test 9	16.0GB 2GB x 2 by 2 in both risers
	Test 10	6.0GB - 1GB x 2 in top Riser, 2GB x 2 in lower riser
	Test 11	6.0GB - 2GB x 2 in top Riser, 1GB x 2 in lower riser
	Test 12	10.0GB - 1GB x 2 + 2.0GB x 2 in top riser, 2GB x 2 in lower riser (1GB modules in first bank)
	Test 13	10.0GB - SAME, but but the 2GB x 2 in the first bank, the 1GB x 2 in 2nd bank for top riser and 2GB x 2 stays same location in lower riser
	Test 14	14.0GB - 1GB x 2, 2GB x 2 top riser, 2GB x 2 x 2 in lower
	Test 15	14.0GB - 2GB x 2 x 2 in upper riser, 2GB x 2 + 1GB x 2 in lower
	Test 16	32.0GB - 4GB x 2 x 2 in each riser

For those of you who would like some actual numbers, rather than pretty pictures, here's the numeric results for the above tests.

Obviously, the 8-core Mac Pro outscored the Quad-core's throughput by almost double. This makes sense; there are twice as many cores to do the processing. What is more significant, though, are the peaks and drops in the 8-core, showing that placement of the RAM does have an effect how fast the data moves through.

Not surprisingly, the best performance came when all four banks were filled with identical pairs of RAM. When the upper and lower pairs match, interleaving gives an additional performance boost.

Also of note were the worst performing configurations - having just a single pair installed in the top riser, rather than one on each riser, like Apple ships them. With this configuration, the speeds plummeted on both the Quad-core and the 8-core models.

So what does this all mean?

While that old chestnut of "More RAM = Better Performance" still rings true, these tests show that by installing your RAM in matched sets of 4 (One pair on top and one pair below), you can get some extra speed out of the deal. In situatons where data throughput is premium, but your budget is small, four smaller modules will ultimately give you more performance than a single pair adding up to the same amount.