Could it have had 40 ROPs instead of 32, or does it need to follow a special format that's dependent on the memory bus - like how the 384bit gtx480/580 had 48 ROPs, the 320bit 470/570 had 40 ROPs and the 256 bit parts had 32?
I see that each ROP on tahiti is arranged in clusters of 4 ROPS, each ROP corresponding to a cluster of 64 shaders. Could this arrangement be changed so that the gpu would retain its shader count, but incorporate more ROPs up to 40 (maybe even 36)?
If not, would doubling the number of ROPs - since that seems to work mathematically- increase the die too much?
I'm asking after hearing speculation that the hd7970 is bottlenecked in this regard, even though each individual ROP has been improved since the last generation.
I see that each ROP on tahiti is arranged in clusters of 4 ROPS, each ROP corresponding to a cluster of 64 shaders. Could this arrangement be changed so that the gpu would retain its shader count, but incorporate more ROPs up to 40 (maybe even 36)?
If not, would doubling the number of ROPs - since that seems to work mathematically- increase the die too much?
I'm asking after hearing speculation that the hd7970 is bottlenecked in this regard, even though each individual ROP has been improved since the last generation.