GPU BLADES Need some factual information !!

[NerdLifeAMD]

So iv been slightly MIA lately. Ill be back in the challenges as soon as a few pressing projects are completed.
Sooo. Ive been working on the crunching rack and its room cooling power ect but im at an impass right at the moment.
Im looking in the general direction of cpu based crunchers and am entertaining the idea of gpu farming.
5800 series cards or around 75$usd here atm. So one of my questions is GPU crunching is not dependent
on a large about of system memory or CPU bandwidth correct? and if so hypothetically a single core chip with say
4 1 giig sticks of ddr3 would create no bottleneck of any kind on the gpus crunching output? and finally from
my other post the speeds of multi pcie slots are confusing me on motherboards. 4 slots (16x-X2 and 8x-X2).....?
what the heck is that ???? and at 8X slot with a 16X card will that hinder the cards performance ?...
I do recall something about when your crossfire the **X speed of the card is cut in half ? which maybe implies
that if i wanted to run all four slots,,,,,, a crossfire would be the only way i could run all of them??????

NEED INPUT go to 10




ok hold on a sec one more thing .......by crossfires very nature
wouldn't simply having a crossfire configuration hinder my output?
I mean they are splitting the load r they not ? they are not sorting twice the load
they are taking a single load and sharing the work ?

[Dirk Broer]

Well, I can see a bottleneck coming when you have a single core CPU with not enough RAM -and 4 GB will not be enough- to support four GPUs running BOINC projects.
As some GPU projects use a complete CPU core to keep it running, I'd advise a quad core CPU in case you want to employ four GPUs on one board.
I'd also advise at least 8 GB of RAM too -and preferably 16 GB- when you're into CPU crunching too. You are running 8 WUs at a time and it's better to be safe than sorry.

Crossfire as such has no advantage in BOINC, BOINC just sees two or more GPUs -provided you have edited your cc_config.xml with the needed line.
It will allocate a GPU WU per GPU it sees, and will *NOT* share a WU over two or more GPUs!

It might help to change your settings in the Catalyst Control Center to have Crossfire enabled as Terry observed.
But when you are satisfied to run them all at stock speed the only thing you should need is the cc_config.xml edit.

Theoretically a card using x16 would perform best and a card at x2 worst, but the effect in real life of this may vary per project.