Mac Pro :: 6GB Versus 8GB Triple Versus Dual Channels/PS4 Tests

I ran some tests on my 2009 Quad Nehalem to try and determine what was up with the triple Vs dual memory "brouhaha".

I posted the results as a new thread because I think it will be useful information for a lot of quad owners, but it was originally going to be a reply to this thread: http://forums.macrumors.com/showthread.php?t=735845

Here we go. Tesselator suggested 3 tests that could show the differences in speed between triple and dual channel bandwidth.

Quote:

As one idea I would maybe try creating a few very large images (16-bit, blank white, blank black, gradient fill) and then duplicating and deleting that layer repeatedly a few hundred times.

So I did them, 10 times each. I could have gone on, but the results were very very stable after the first 2 attempts.

Set-up: a 40Mpx, 16bit image (8000*5000). First test it was simply filled white; second test: black; and third test a black to white gradient. I added a fourth test, using a real (photo) 12Mpx RAW image from my Nikon D300.

I created (took a while!) an action with 350 repetitions of "duplicate layer" and "delete layer", followed by a red fill to let me know the action was done. The same action was used in all four tests.

The computer was restarted before each of the four tests, which may explain the irregularities on the first 1-2 attempts. Nothing else but PS4 was launched.

The results are interesting:

We can clearly see that the simple white and black fills show a speed difference of around 10%.

We can also clearly see that this difference disappears when a more complex image is used. The use of more complex images represents a much more realistical use of PS.

To make things even more realistic, I also tested RetouchActions's speed test on my own 12Mpx image. I use nearly all of the operations of that action on a daily basis, so it's a lot more representative of the work I do on PS.

Here are the results:

The results are clear: 11% increase in performance using 8GB of ram (Vs 6GB) when working on a 12Mpx image.

Added info: number of page-outs after running the 10 test series (after about 45 minutes of intense PS work):
-17K when using 6GB (1.7K page-out avg).
-10K when using 8GB (1K page-out avg).

For me the results are definitive: unless I plan on working only with full black or full white images (not even black and white!), having 8GB is better, even when working on smallish 12Mpx files. I imagine the differences would have been even greater using bigger file sizes of actual complex images.

What would now be interesting: someone with a 2009MP Octo doing the same tests at 12GB and 16GB.