If the first 20 nm chips would show up in late 2014 rather than early 2014, then I thought there would be three 28 nm series of chips at a ~1 year cadence, even if the third one is a hybrid like Northern Islands or some type of stopgap. If there really are only two 28 nm series, then I wonder what motivation (if any) they had for "skipping" the late 2012 / early 2013 series instead of a late 2013 / early 2014 series.
I think NVIDIA's 700 series (minus the Titan, although that may not be named under the "GTX 700" series) will end up similar to AMD's 8000 series (at least what we've seen so far), in the sense that there won't be significant performance increases. From what I've read, the 680 appears to have a memory bandwidth bottleneck. So if the 700 series GK104 parts stay at 6 Gbps maximum memory speed, they may not have significant core clock speed increases over the 670/680. If they move up to 7 Gbps (probably less likely, maybe that'll happen with GK114/GK204, if at all) then would open the door to up to similar core clock increases. Maybe the successor to the 660 Ti (and 650 Ti) won't have reduced bus widths.
If the 680 successor has similar or identical specs to the 680, NVIDIA might reduce the TDP slightly (or maybe not—AMD didn't seem to with the 8970 etc.), which may allow a dual-chip 690 successor to be two full non-clock-reduced 680 successors. However, if the Titan is as good as what certain rumors claim, there might not be a direct 690 successor.
Also, I don't think a Kepler refresh in late 2013 means that there won't be 28 nm Maxwell in H2 2014. There's a fair amount of room above even a refreshed GK104, assuming they don't add on SMXes to it or increase its bus width, for a Maxwell chip to slot in.
