Quote Originally Posted by BiFfMaN View Post
this

a fairy showed up?

Quote Originally Posted by Leeghoofd View Post
Insulation was time consuming :p , a team member is trying to get the code knackered to use it on other boards (but the pci spacing could be an issue) but maybe guys like TIN or co might be better at that
insulation: but it worked out in the end? k...
slot spacing: just use flexible pciE risers...
code? what code? why do you have to lay hands on any code?

Quote Originally Posted by fgw View Post
each slot uses 16 pcix lanes, so all cards are capable of using full 16x bandwidth for communication to an other card.
mhhhh no, each pair of cards can talk to each other through 16 lanes, when the pair wants to talk to the other pair they talk through 8 lanes, and to the nf200, which talks to the IOH, which talks to the other nf200, which then knocks on the door of the other pair
thats why nvidia actually recommended all board partners to piggyback dual nf200 chips, so the hops stay low...

nf200_1 connected to vga_1 and nf200_2 which is connected to vga_2 and vga_3

of course back then it was all about tri sli...
but even for tri sli this config made no sense because hops between cards dont actually matter as it turns out... cause despite what nvidia claimed, the inter card communication doesnt really need a lot of bw and low latency...

so what really matters is bandwidth to the cpu and low latency to the cpu... and for the first the nf200s dont do anything, 8 per card stays 8 per card (bandwidth wise), and for the latter they make things worse...

the only reason there are nf200s at all is cause nvidia wont allow 4way sli without it...