Hmmm...has anyone ever thought that they might split the 300 million transistor G7x type core up into two seperate 150-200 million transistor cores for the RSX, each with it's own 128-bit memory interface. In essence, they would be preserving the 256-bit memory bandwidth (because the lesser throughput from fewer ALUs on each core would no longer be a bottleneck for a 128-bit bus), increase yields, and lower costs....and in the end they would be able to shrink the chip in the future on a 65nm or 45nm process...
Maybe I'm crazy...but why wouldn't Nvidia thought of this? I'm sure they have...maybe that's what all this dual-core rumor rumblings have been about.
Can any devs give me a wink or something?
This would be a very interesting step for Nvidia to take and a welcome suprise