Conclusions
Well, in a twist from normality, this article is about speculations. This means that, while I wrote reasonable things, I can not be 100% sure about what GK100 / GK110 reserves to us.
However, I really think that the last hypothesis is quite reasonable: a 2048 ALUs, ~500 mm2 chip should guarantee very high compute performance (2X SP/DP rates compared to Fermi, clock to clock) and significantly higher gaming performance (48 ROPs, 384-bit memory bus).
Obviously, even if this prediction proved correct, an even greater questions is the final clock speed of such a beast. If history is going to repeat itself as with GF110/GF104 transitions, it is possible that GK100 / GK110 clock will be only slightly slower then GK104, perhaps in the range of 900-925 MHz for the base graphic clock. This can translate into ~25% advantage in gaming workloads.
Time will tell whether these predictions will prove correct or not. If you wish, you can write me at
Have a nice day!