I could be wrong with this, but for the past few days I've been reading and reading and researching and reading research and researching reading and... well, you get the point I think.
Anywho, what I've come to realize (and this is the part where I might be wrong) is that in order to get that magical 21GB/s of memory bandwidth, any memory you plan to install in the MacPro
MUST be in multiples of 4.
The reason I say this is because if you look at the way Apple prices and parts out the RAM on the order page, it looks like this:
1GB (2 x 512MB) (stock configuration and it's just not really enough)
2GB (4 x 512MB) [Add $300]
4GB (4 x 1GB) [Add $1100]
8GB (8 x 1GB) [Add $2500]
8GB (4 x 2GB) [Add $2700]
16GB (8 x 2GB) [Add $5700]
Now, aside from the measly 1GB they do stock (and really, I think Apple should have started at 2GB and never offered the 1GB at all, possibly as an upgrade maybe to the stock) but just 1GB stock alone? Man that's just ridiculous for their high-end "pro" machine knowing how expensive this stuff will be for us end users.
So, looking at the other choices, you can plainly see a pattern there: they're selling the sticks in configurations that are multiples of 4. I found some article a few days ago amongst the hundreds I've looked at and it was the only one that hinted to this seemingly very important aspect of buying memory for the MacPro and now for the life of me I can't find the article or the bookmark to the page. Hell I spent the past two hours just redoing all my history (and that was a LOT of stuff in the past week regarding MacPros) and I simply cannot find that specific piece of information again.
If you choose to get 2x<etc> sticks, 512, 1GB, 2GB, whatever, you're hamstringing your performance and you're just not going to get the full 256 bit wide memory bus really screaming along. I don't think these sticks work with data in 64 bit wide chunks, so again, I'm just offering my point of view based on some simple math, I guess.
Multiples of 2 (2x512, 2x1, 2x2) do add more RAM to the machine, obviously, but in the end using multiples of 2
just might be killing off a considerable chunk of speed for memory intensive calculations and number crunching.
When I get my MacPro I intend to spend the first few days benchmarking it three ways from Sunday, with OSX, XP 32 and 64 bit (yeah, I know the 64 bit version is effectively dead in the water, but I've got lots of time to kill), and Vista 32 and 64 bit. Might even throw a few Linux distros on it if I can.
Wish I could get my hands on that Leopard preview from WWDC, but I can wait.
In light of this, and again I know I might be wrong, completely off my rocker, but I swear I read some benchmarks of a MacPro that was tested with 2x<etc> configurations and then 4x<etc> configs and the memory speeds shot up dramatically because they were installed correctly in multiples of 4.
Hopefully someone will either flat out say I'm wrong and back it up with hard data, or I'll do it myself when my MacPro arrives hopefully sometime before next week is over - I'm ordering it on Wednesday when I get my hands on some cash I have on hold at my bank till it clears at that time.
While I wish I had
more money than I actually will, I'm bumping the stock memory to 2GB and getting that 4x512MB setup. That way I can yank two off the boards and see if my wacky off the wall theory has some weight to it. It'd be nice if I'm right.
Thanks, and I hope I at least provided something for people to consider...
bb
ps
What would be great is if someone out there bought the stock 2GB as I'm doing, from Apple, as 4x512MB. And then after receiving their MacPro they bought an additional 2GB as 2x1GB for a system total of 4GB of RAM. This would make the perfect platform for testing my theory, and something I simply am unable to do.
Remove the 2x1GB sticks and do a memory bandwidth test of some kind, under Windows you have several choices: PCMark05, Sandra07, etc etc, but under OSX I guess Xbench would work in a pinch. Run it three times, get the average score then shut down.
Pull out the 4x512MB sticks, put in just the 2x1GB so you're still at 2GB total, and rerun the benchmarks. If my theory is correct, you
should see the memory bandwidth drop through the floor with 2 sticks installed instead of 4 sticks wide.
I hope to test this at some point myself, but for now, I'm only going to have access to the 4x512MB that Apple is going to install. I know that I can do a 4x512MB test and then do 2x512MB but that's not a fair comparison really because it's not the same amount of RAM in the system like 4x512MB vs 2x1GB is.
Last question: Does anyone know specifically what brand of memory Apple is using for the stock RAM, and possible part number information about it? Thanks in advance...
pss
AHA!!! I think I found what I was talking about. On the MacPro order page, in the RAM section, if you click the "More info" arrow for RAM, one part of it says this:
"Mac Pro uses 667MHz DDR2 fully buffered ECC memory, a new industry-standard memory technology that allows for more memory capacity, higher speeds, and better reliability. To take full advantage of the 256-bit wide memory architecture, four or more FB-DIMMs should be installed in Mac Pro."
Now, as for whether or not it's specifically in multiples of 4, I have no idea, that's what I intend to test out unless someone beats me to the punch. If it's true, and you do need to install in multiples of 4 sticks, that's a pretty significant piece of information I'd say.
Have fun, always...
