WEll, obviously you don't. There is no differences between the Gk104 and the Gk110 apart from the fact that the Gk110 has much better compute capablility. IF you read the white paper on the card you would see that.
It has the exact same SMX's with 192 cuda cores the only difference is
the GK110 has 64 FP64 cuda cores.
The only changes that have been made have all been to make it a better HPC card.
There is nothing about the GK110 that suggest it's a second generation Kepler. Nothing.
The DP performance isn't a minor change, its a big change, and what suggests its a second generation kepler is that people talk about generations like its a huge thing. The "second gen" Fermi's were not much more than respun first gen ones, thats basically it, IE " we made a boo boo with our design resulting in lower yields and higher power, this is a "fix", not much else". Thats what the 580gtx, thats what the GK110 is, its a "fix" to GK100, which was so bad that it didn't even get a release like the 480gtx.
GK110 is as much a second generation kepler as the 580gtx was a second gen Fermi, ultimately its semantics, one person calls it a new gen, another person calls it a respin to improve yields, another person calls it a respin to fix the horrendously yielding first version... all the same thing here.
A Fermi 580gtx was barely different to a 480gtx in anything but yields, and a GK110 will be hardly any different to a GK100 in anything but yields most likely. Any major changes will happen for 20nm, this is a refresh, as the 580gtx was, in the general usage of the term generation.... the 580gtx wasn't a second gen Fermi and the GK110 isn't a second gen Kepler, they are just both fixes/slight tweaks mostly based around yields.
EDIT:- also, the GK110 wasn't built from the ground up to do compute, its the same basic architecture, these things are completely joined together, and there isn't a split from Nvidia in anything but marketing. Last gen the GF104/114 from what I can remember, had the same DF performance(at least on die, if it was lowered in drivers) as the high end parts, this gen they actually stripped parts to increase DF performance fully out of the lower end chips to save die space and power.
If anything I'd call the GK104 the consumer tweaked card, the GF104/114 were simply smaller versions of the top end chip with no functional difference, aside from marketing, and maybe drivers, they are essentially the same core.
There is more of a split this gen, to reduce die size, improve yields, and profits, because there really isn't much need for DF performance for home users, but this is a fairly basic tweak, most of the chip is the same architecture, same shaders, same rops, same everything. In no way can you call Gk110/104 chips built from the ground up for different things, its one design with a modular part to it where Nvidia can choose to add in some DP boosting parts.
Larrabee, or whatever the hell the latest one is called, is built from the ground up for compute, the GK110 isn't, it would indeed have been on sale as a consumer part, blowing AMD out of the water, if they could produce it in any sensible numbers for sensible prices, yields stopped it being sold as a gaming part up till now, nothing else.
