Quote Originally Posted by Cybercat View Post
Whether or not this is true, it would make sense to do, right? It would save on PCB costs significantly.
Keyword is probably PCB length instead of costs, though.


I think this would make it up nicely in a dualcore situation.

------------------------------------------------------------------------
GPU1 <-> GPU 1 Sideport Interface <-> GPU2 Sideport Interface <-> GPU2
------------------------------------------------------------------------
|
|
|
V
PCIe 2.0 x16

Aka GPU1's sideport acts just like the PLX chip (on the 4870X2 they act like extra traces)

If they have a crossbar approach it would be even more interesting to watch.

Top boss of Chiphell (pretty much always accurate) is implying that memory will be shared between the MCMs.