As i understand it it's not for quad SLI but it allows you, with the right motherboard, to SLI a pair of them and have a 3rd which will handle the physics.
No actually it's for unbound chains.
The original SLi on the GF6 & 7 only allowed 2 way communication along 1 path, thus the GX2 was not really quad SLi, but SLi+SLi, that was actually a limitation on the chip. there was no such limitation to ATi's solution (nor their previous solutions like the E&S simfusion [which allowed up to 32sets of cards in a rack mount system]).
Now neither the R600 nor the GF8 are limited, and quad SLi/Xfire are quite easily possible whic true point to point communication between the chips (without going through the PCIe lanes). This also allows faster card to card communication by doubling the throughput. However moreoften than not these connectors aren't really saturated, they are mainly used to control the cards.
Now while ATi has mentioned support for Quad Xfire, and nvidia has already demoed their bastardized version of 'quad', the interesting concept is the idea nV was mentioning of asymetric SLi involving 4 cards, but 1 for physics, and 3 for graphics, and then the ability to adjust tha based on needs. So no physics, quad SLi, some physics, tri (asymetric) SLi, heavy physics, dual Physics cards, and dual graphics.
Alot of it is still sketchy, with only leaks and such, but the gist of it is that now there are no physical limitations for SLi/Xfire in greater numbers tha 2, and while this may be big for us, the largest impact may be in the workstation and GPGPU fields, howvwer there's been little said there too, and I doubt we'll see any of it until the R600, and GF8900 and G80-GX2 variants hit so the competition can size each other up. Also the drivers are likely far from ready for both companies, we know the current state of VistsaSLi, and I doubt Quad for either company would fair much better at first, but I sure would love to see some serious use of that super tiling with more than just 2 cards.
I thought current mobos were limited to 20 pcie lanes. which is why dual x16 slots really only run in x8 mode? So single x16 slot and some x1's. Else same but 2 x16 which run x16 when only one slot is used. X38 chipset due this fall with PCIe 2.0 is supposed to have dual x16 running both at x16. (and twice the data rate at 1.0 for four times the bandwidth running to your SLI/Xfire setup) does 680i do true x16 SLI/Xfire now?
Both solutions could easily provide 4 x 8 lanes, and with the next generation will add to that with the expectation being higher numbers with the desire to be 16+16+16+8, but early board focusing on physics with the 16+8+8 and 16+16+8 configurations. The 4x8 is already on the map;
"When pressed further about why two connectors were used, ATI asked us to envision a system with 3 or 4 graphics cards installed. With 2 channels, cards can be easily chained together. This does offer ATI a little more flexibility than NVIDIA in scaling multi-GPU configurations, but it is also a little more cumbersome and offers more small parts to lose. Overall, though, the 2 channel configuration is a good thing."
Just a thought but since there are two SLI connections, say you want to do normal SLI with the 8800GTX, if you use an SLI bridge to connect them to each other can you use two SLI bridges? would that increase performance at all, the way i think it would is because the cards are now "doubled linked" sharing information could be a lot faster.