Soleron said: "GCN 2.0" is just a rebrand, there's no difference. R500 is also unrelated to R600, it was a cancelled GPU.
Anyway when the consoles come out I'll post on your wall whether I was right or not. I'm confident.
|
R500 is VLIW-5, it's exactly the same as the R600 architecture without DX10. The codename has nothing to do with the architecture in this case. R500 is based on R600 VLIW-5 architecture. The custom part is not the architecture but AMD reduced the # of functional units, which made it worse, not better. It's really just a VLIW-5 240 SP HD2900 (or 48 ALUs, 5-wide vector units (VLIW-5)). The GPU clock is dropped from 600mhz to 500mhz because it was fabbed on 90nm instead of 80nm like 2900GT. That architecture has ROPs tied to memory controllers. This is why to lower power consumption, they chopped off 8 ROPs and with them half the memory bandwidth was wiped out. HD2900GT had 16 ROPs and 256-bit memory bandwidth and that's why R500 only has 8 ROPs and 128-bit memory cuz it's gutted. That's what I am saying if you look at the specs of HD7970M/8970M, that's the fastest you can ever have without going cross-fire. I am sure they'll have eDRAM to help, but I personally think the fundamental functional units (shaders, TMUs/ROPs, memory bandwidth) of the GPU will be HD7970M/8970M at most. I hope they keep 2GB of VRAM though.
Anyway, if you are suggesting that the GPUs will be even more advanced than HD8970M, I'll be pleasantly surprised and would glad to be wrong on this one :)
I think MS might stage a separate event for launching the next console instead of doing it at E3. What do you guys think?