This is the smallest perf. increase in a flagship from one node/arch to the next I have ever seen nvidia deliver. It sucks for me personally, as a typical 50% nv increase would give me more perf. from 2 cards than from the ones I have now. These cards are just like 7970s to me, I'd need to overclock 2 stock cards for more perf. Without any guarantees of getting the clocks I'd want, it's not worth it. Never mind that the perf. increase at 1600P for the 680 is looking to be lower than it is at 1080P.
True. GTX680 is a disappointing flagship generational increase for NV. I still think this is a re-badged GTX660Ti/GTX670. As I noted many times already, Nvidia's next generation mid-range card has always either matched or beaten its previous high-end. In this regard, GTX680 is much faster than a historical mid-range next generation NV card (so it prob. should have been called GTX670Ti in fairness).
Having said that, it was obvious starting with the entire HD7000 series that 1st 'half' of 28nm generation is likely going to be an expensive generation. HD7770 was the biggest pile of **** I've seen since GTX550Ti launch. We have heard for months that 28nm yields were poor, TSMC was wafer supply constrained due to lack of Fab capacity, that they weren't expected to ramp up production volume significantly until Q3 2012, and that the concurrent demand from other customers (Qualcomm, Apple) meant much higher prices per wafer than during 40nm transition, etc.
If we stop focusing on GTX680 vs. HD7970 for a minute, take a look at GTX680 vs. HD7950. GTX680 is beating HD7950 by 30-40% and the 7950 is a $450 card. That in itself tells us the entire HD7000 series generation is mispriced. HD7950/HD7970 were overpriced to begin with from a technological price/performance curve (i.e., improvement would have been HD7970 performance replacing HD6970 at $369). From a business case, their pricing made perfect sense though since AMD just took advantage of early adopters, enthusiasts or people who don't follow GPU cycles.
NV is really laughing here. They are about to sell us a $500+ 294mm^2 die chip with 256-bit memory interface and 2GB of VRAM. No wonder NV breathed a sigh of relief when they said Tahiti's performance was underwhelming. Think about it, if NV's mid-range tends to topple their previous high-end chips, then prob. NV intended GK104 to be a mid-range card at $300-350 or so. If GK104 was meant to be high-end, where is the rest of the desktop Kepler line-up? Nowhere. It's not launching this week. That's because GK104 is just a mid-range card and it was never intended to be cut into 5-6 SKUs; that's why they aren't ready. Could you cut GTX460 into 5 smaller SKUs? It's no wonder NV has none of them ready for launch. Where is the GTX660, GTX660Ti, GTX670Ti, etc.?
This smells like NV took a GTX460/560 style chip, spent a quarter waiting until 28nm yields improved so they could launch a "factory pre-overclocked" version (hence initial clocks of 706mhz -> 1006mhz), rebadged it as a GTX480, scrapped the real GTX480 entirely since they couldn't get it out on time (for whatever reason) and because AMD basically failed to deliver this generation, NV is about to get away with selling a GTX460 at $500!
Now the "real GTX480" will become a rebadged GTX780 sometime later (maybe even Q1 2013), and because GK104 was meant to be mid-range all this time, NV has no choice but to use GTX560Ti/570/580 cards in the interim as their mid-range and upper mid-range cards. Perhaps that's why they are still not EOL. If Nvidia launched 2-3 cards based on GK104 this week, didn't neuter GPGPU compute and outperformed GTX580 by an average of 50%+, I would have believed that GK104 was the real flagship.
That's just my wild theory.