Different versions of Everest will give totally different results. So I do not see why the test isn't conclusive. I only wanted to show that the mixed mode is just as fast as identical DIMMs in three slots of three channels.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
32gb of RAM in 2009 Quads!
- Thread starter kxfrog
- Start date
- Sort by reaction score
You will have 18% better performance with the Quad with the vast majority of applications. I believe there's only a handful that need or take advantage of all 8 cores. It depends on your usage. There's a thread over on apple.com with some poor pissed off soul who bought a 2.26 octo-core assuming it would outperform a 2.66 Quad based on the benchmarks. Well, benchmarks don't tell the true story. Now he's so upset... LOL! With today's Mac Pro and the current state of multi-core aware applications, you really need to examine your usage patterns to determine the best configuration that meets your needs.
Until you can verify that the test is reporting dual vs. tri-channel performance correctly, I don't know how you can be certain of the test results. What if your memory controller is borked and running dual-channel all the time? Your speeds do seem a bit low for tri-channel DDR3 at 1066.
As for the differences between our two tests... I suppose a difference in test versions could explain it, but the Everest memory test is pretty solid and as far as I can tell, doesn't change much (at all?) from version to version... It might also be the octo vs. quad (and/or QPI link)... who knows...
I guess I should try the test again with the same version you are using. I've got a new SSD that I can install Windows on temporarily before I integrate it into my OSX array... I'll try that over the weekend.
Not really, as there's precious little software than can actually utilize triple channel's bandwidth. Dual channel is fine for most anything.
Absolutely right... there's little to be gained from this.
I'm just curious to resolve the ambiguity surrounding how the IMC handles four populated DIMM slots. Gugucom's theory, if proven true, would be a nice bonus.So, I spent some time this afternoon installing Win7 on my spare SSD and running some memory benchmarks with Everest and Sisoft Sandra.
It turns out that Everest is useless as it appears to only test dual channel... running the benchmark on a tri-channel configuration resulted in the exact same scores as the dual-channel run.
@Gugucom, if you can run some tests using Sisoft Sandra's memory benchmark, that would be great... it appears to be able to actually utilize a tri-channel setup.
So I posted benchmark results in single, dual, and tri channel modes (removing sticks of ram from my 3x2GB setup as required) below for Sisoft as well as the dual/tri channel results for Everest just to show it's useless.
Attachments
I have done some more tests. Both Everest (latest Version) and Sandra do recognize each memory config correctly and report dual or triple channel operation also correctly.
In Everest I do not get lower results for running with dual channel only. Same result that you got.
In Sandra I get:
25.77 GB/s in mixed mode, triple channel
24.07 GB/s in 2 GB only, dual channel
24.17 GB/s in 2 GB only, triple channel
In Everest I do not get lower results for running with dual channel only. Same result that you got.
In Sandra I get:
25.77 GB/s in mixed mode, triple channel
24.07 GB/s in 2 GB only, dual channel
24.17 GB/s in 2 GB only, triple channel
A few things confuse me about your results...
- Why are your dual and triple channel results identical?
- Why are ALL your results higher than a tri-channel DDR3-1600 reference system?
What were your memory configurations in each of those tests? Can you post screen shots like I did? I'd also like to see what you were getting in Everest as a sanity check.
Something is not right.