Of course, our major concerns surround whether or not Nvidia can continue achieving successful 28nm yields, maintain transistor leakage to very minimum levels and of course, stay within respectable selling margins without getting too greedy on this computational beast of a Kepler chip.
It is expected that GeForce Kepler GK110 may have up to 2304 CUDA cores - up from GK104's 1536 CUDA cores - thanks to its bigger die which allows more SMX units (Streaming Multiprocessors x 2) to be placed on silicon. The new chip will also contain a total of nearly 6 billion transistors all drawing a total TDP between 250 and 300 watts.
In terms of raw power, a single GK110 GPU is expected to have up to 4.50 TFLOPs single-precision compute performance. In comparison, AMD's Radeon HD 7970 flagship single-GPU has 3.79 TFLOPs, the GeForce GTX 680 (GK104) has 3.09 TFLOPs and the previous generation GeForce GTX 590 (2 x GF110) dual-GPU card has 2.49 TFLOPs.
All in all, having three GK110s in Triple-SLI would still give you more single-precision performance at 13.50 TFLOPs than having four GeForce GTX 680s (GK104s) in 4-way SLI at 12.36 TFLOPs. In comparison, having two AMD Radeon HD 6990s in Quad-CrossFireX (four Antilles GPUs) would only give you 10.2 TFLOPs.
Nvidia has yet to reveal a name for its flagship single-GPU card, but we have two theories. Either the company will name it GeForce GTX 685 and call its dual-GPU the GeForce GTX 690, or it will decide to go with GeForce GTX 685 for the single-GPU card and call its dual-GPU the GeForce GTX 695. Nevertheless, the latter nomenclature hypothesis would line up with Nvidia's GeForce GTX 285 and GeForce GTX 295 dual-GPU cards from its GeForce 200 Series desktop lineup.