I can't help but wonder if the release in Q108 (G90?) is actually just a 65nm G80 with hella higher shader clocks (amongst faster core clock/memory).
A g80 with 2400mhz shader clocks...920 Gflops if mul exposed, 614 if not, ...Pretty darn close to a teraflop, and would mesh with "almost a teraflop".
Surely would be interesting if that's the way Nvidia is planning to scale the architecture.
If it's something like 160 SPs (as some have speculated, and perhaps 512-bit bus) with such shader clocks, that thing would be the mutts nuts with literally TWICE the shading power of G80 (if shaders were at 2400mhz).
g80 = 128x3x1350= 518.4 but as mul is not exposed = 345
?? = 128x3x2400 = 921/614 (w/-w/o mul) - logically possible.
?? = 160x3x2400 = 1152/768 Gflops - just plain loony.
The former sounds more reasonable (easier to create/profitable) though.
Reply With Quote
Bookmarks