Quote Originally Posted by SKYMTL View Post
I am guessing that's 16 lanes from the NF200 and 8 from the PCH but I could be mistaken.
PCH doesnt have any pciE 2.0 lanes, only 1.1 and it DOES have 8 lanes iirc but only 4 of those can be bundled to a 4x link...

Quote Originally Posted by LardArse View Post
slot 1 will always be native so that single card you dont have to go thru nf200.
u can run sli or cf on 1,2 or 1,3 its exactly the same.
thats still really confusing... the cpu only has 16 lanes, so how can you use nf200 AND have the first 16x slot hooked up to the cpu directly?
unless you route 8 lanes to the first 16x slot and the other 8 lanes to the nf200...

so then its:
8/16
0/16
0/8/16
0/8

very nice!
tri sli will be a bit ackward then, running 8/16/16 hahah
wow there are so many possible combinations... interesting!!!

single vga:
1.) 16x to cpu (slot1)
2.) 8x to cpu (slot1 with slot 2/3/4 occupied by any other card)
3.) 16x to nf200 and then 8x to cpu (slot2 or 3)
4.) 8x to nf200 and then 8x to cpu (slot4)

you already said 1.) is slightly faster

dual vga:
1.) 8x to cpu + 16x to nf200 and 8x to cpu (slot1 + slot2/3)
2.) 8x to cpu + 8x to nf200 and 8x to cpu (slot1 + slot4)
3.) 16x to nf200 and 4x to cpu + 16x to nf200 and 4x to cpu (slot2 + slot3)
4.) 16x to nf200 and 4x to cpu + 8x to nf200 and 4x to cpu (slot2/3 + slot4)

according to nvidia 3.) should be the fastest, but my guess is 1.) is faster...
and the best way would be 1.B) using the 1st and 3rd slot so the 1st vga has lots of fresh air and you have the pci slot free... very nice!

tri vga:
1.) 8x to cpu + 16x to nf200 and 4x to cpu + 16x to nf200 and 4x to cpu (slot1+slot2+slot3)
2.) 8x to cpu + 16x to nf200 and 4x to cpu + 8x to nf200 and 4x to cpu (slot1+3+4)
3.) 16x to nf200 and ~2.5x to cpu + 8x to nf200 and ~2.5x to cpu + 8x to nf200 and ~2.5x to cpu

3.) shouldnt be all that bad according to nvidia, but i think the cards will be highly bw starved...
1.) is the fastest i guess, and the first slot can be a 295 and the 2nd and 3rd slot run through the nf200 so 2 gtx285 cards can be masked as a second nf200... very nice
2.) is a nice option if the cards are slim or watercooled as the pci slot is still free then and the first vga has lots of room for fresh air.

quad vga:
1.) 8x to cpu + 16x to nf200 and ~2.5x to cpu + 8x to nf200 and ~2.5x to cpu + 8x to nf200 and ~2.5x to cpu

Quote Originally Posted by SKYMTL View Post
I guess it all really comes down to the possibility of future GPUs saturating a native x8 interface.
in my testing even dualgpu cards dont saturate an 8x 2.0 link...
the perf loss from 16 2.0 to 8 2.0 lanes for a dual gpu card is there but its tiny...

i only tested with a 4870x2 though... nvidia usually likes extra bandwidth more, so there quadsli with 2 295 cards might see a drop when running them on 8x/8x... but since the board will have 4 slots to run quadsli with 4 cards... that shouldnt really matter...