Quote:
I think you guys are judging in the photos mislead by the architecture. It seems to be that for the shader units have been lined up to two instead of single row, but are cmq intercommunication between them, and not only speaks with block 1 block 1 and block 2 speaks only with block 2. I think this is only a constructive solution "geometric" to optimize the molding of the chip, so this scheme to "butterfly" should not take you on flights of fancy too and I think dual GPU. Not counting that, and they all end going into the UDP (ultra dispace processor). Then you see as with R600 on it is he who takes care of sharing the work among the various sub array of microcore. Hence, for him 22 or 22 in a row numbered from 0 to 21 or have 2 lined up to 2 always between 0 and 21 is simply a physical movement of constructive optimizations.
Different speech but if the calls made to or from UDP were like 0-10 or 0-9A and 0-10 or 0-9B at that point but we could speak of architectural style of dual core "pure" because the calls are mirrored in 2 different units then A and B.
First we must need to see if the calls have to be mirrored, so if each unit 0A must necessarily do what the 0B and that means that the worst results of architecture of rv870 will result a 800x2 more than a 1600 SP.
I do not think so I think that they are independent.
Atlas, then you just see how they are called arrays. If there are several contiguous, then it is dual core, dual core is only for the geometry construction. If they are considered 2 separate units well then you endorsed the solution "dual core"
