Yep essentially the shaders can do all full 3 operations rather than the 2 previously.
Also, it's not 80 ROPs, its 32 ROPS.
Like G80 architecture, its 4 ROPS per 64-bits of memory bus hence 512-bit bus -> 32 ROPS, 448-bit bus -> 28 ROPS
And it looks like the ratio is 24 SP's per 8 TMU's as well which isn't quite the 2 to 1 for G92 but better than the 4 to 1 on G80 (though it was 32TA, 64TF)





Bookmarks