Just a couple of points though - I thought GF110 as more a manufacturing and optimisation improvement than a computer/graphics compromise closer to graphics. I might be wrong.
You are right. GF110 didn't drop any compute feature or capability. It's GF104 and below that did it. They simply made it work as it should with GF110.
To counter Ben though the original rumours suggested a bottom up launch for NV to get the process right. The last time I checked, the Kepler behemoth gpu isn't scheduled to arrive until 2013. GK 104 is the card for now (a high end but not top end single card).
Oh I didn't pay attention to this paragraph. You are not completely countering me. GK104 is the card that is going to release "now" and that's what I've been saying. I wouldn't say end of February is now tho (just like I would even question if Tahiti actually launched yet), especially considering that, hard launch of not, they will not have sufficient cards to meet demand by a long shot. I remember the days in which a release with less than 250k cards worldwide was almost considered a paper launch. Now 10k cards on US only and people take precautions not to say paper launch. Bleh, IT IS paper launch, of course it is.
Regarding "Kepler behemoth", show me the proof that is going to release in 2013 and please don't link to that chart that screamed fake (i.e when has Nvidia EVER given dual-GPU cards a codename? Fake. Only AMD does that.), and don't mention Charlie D. and associates who still think that no next gen Nvidia chip has taped out yet and hence say it's 2013 launch (if all goes well!).
As I see it, Kepler performance is going to release in Feb. and Kepler behemoth as you called it, will launch in April, probably later. Show me any evidence that demostrates that Nvidia can't release their behemoth*, namely GK110, in April-June timeframe. And again refrain from using the usual suspects who still think that no Kepler chip exists yet or that for example only 9000 waffers of GF100 were ever made (absurd considering the amount of cards they sold), for example.
There's just simply too much confusion regarding what Kepler is. Some people refer to the high-end chip as Kepler, which will launch in April or later. And hence are calling GK104 a stopgap card, because it's maybe not completely the same architecture as the high-end chip. But that's simply because they have been spoiled by how AMD uses codenames. Some other people rightfully include GK104 and hence report Kepler in Feb.
* How do we know it's going to be a behemoth anyway. I'm talking about ca 500 mm^2 behemoth like Fermi. They don't really need it this time around. First and foremost it will be a 512 bit card, hardly more than that. That's only 33% more transistors devoted to that part of the chip, ROPs and L2, included. MC+ROPs+L2 is almost half of the chip in GF100. They don't need a massive increase for Kepler high-end in that front, especially if they sorted out memory clocks. Then it really depends on how they arrange the SMs. For example and it's just an example, GF104/114 is a 1.9 billion transistor chip, 384 SP, 256 bits, so in theory they could simply mirror it and end up with a 768 SP, 512 bit chip. Any doubts about such a chip decimating the GTX580 and by extension also easily beating Tahiti? Well remember 1.9 billion *2 = 3.8 billion transistor, a fair bit less then Tahiti and hardly a behemoth. Not a fair comparison because GF104 lacks the compute capabilitites that high-end Kepler will certainly need. But between such a card, and another hypothetical 500 mm^2 and 6 billion transistor behemoth, there's a world of a difference and 2 billion transistor budget to play with. Nvidia in theory and if they don't go nuts with compute over what is already sufficient, could release a card with 2x times the arithmetic capabilities while staying well below the 500 mm^2 mark. Will they do it and stay below? I don't know it was just pure speculation.
Seriously. Time to reconsider the form factor. Make the "PC" a huge GPU environment... and make the CPU an add-on board. ATX form factor... nV style.
