All the pricing for AMD and Nvidia can traced back to early in 28Nm coming into the picture. Sorry this is Long Winded.
TSMC – Indicated early on 28Nm production would be more costly... "by somewhere between 15-25%"
Nvidia – Would start purchasing full wafer production, not per chip as had been the case previously.
AMD – Foresaw GK100 being expensive given the die-size, but then when it was cancel AMD assumed they had cornered that market. Then they received less than optimum Tahiti's from TSMC production, but still figured they could go for brass ring asking $550.
Nvidia – Finds they can have GK104 best Tahiti with the implementation of Boost, even if that adds cost they can easily under-cut a 7970 @ $550.
Nvidia – Surely designed Kepler knowing they'd buying wafer production, and designed the whole "multiple variants" from a single wafer (not 2 and then one gelding as AMD traditionally). They've binned 5 different variants from every GK104 wafer, which absolutely gives them a price advantage. That alone was the early strategy that really turn to be the boon for Nvidia against AMD LE/XT harvest.
AMD – Tahiti doesn't have the efficiency to offer higher clock performance. I always thought Tahiti XT was to be a GHz offering, but at first because of TSMC production issue they released with 7.5% lower clocks believing good production could provide a straight-up 1GHz card. Even once good production was the norm, AMD found going with TSMC "HPL" (lower power) process and GCN architecture wasn't working out as well as first wafer trials might have indicated and took a page from Nvidia also adopting the Boost approach.
Nvidia – Kepler architecture was from the beginning designed to be extra efficient so it could profit more from the "HP" without being inefficient. The problem was at first GK100 was still a unwieldy. I think Boost was originally something conceived/experimented to aid the GK100 although quickly realized Boost could make the GK104 a contender to Tahiti. The cancelation of GK100 was because of realization for Boost performance worked on the GK104 (also believe due more to cost, maturity in the TSMC process, but also a fix or two). So canceling and delaying things a little help; lengthen time for good HP GK104 wafer production, while time to develop/implement boost. That was another great boon to Nvidia.
That might be how the cookie crumbled… just one person’s thinking. So AMD had miss-steps, but assumed they would come out on top... So why underprice what traditionally Nvidia had always shown to do? Nvidia had done things smarter, but had they not found some windfalls they might have had to do things different. So Nvidia came to the table in a healthier position than ever, and pull the rug out from under AMD. I'm fairly certain Nvidia could move further on price when AMD made cut and added bundles, but Nvidia needed to hold firm to maintain pricing, so when stacking the 780 and Titan on top of the revamped 6XX cards pricing appeared apropos down the road.
Its the way.
