On the immediate reply to the first query of the figure I said it was approximate... get over it already...
Implementing the original DX10 specification on the 80/90nm process at the time would have hugely pushed up costs as it would have required quite a bit of extra hardware development, pushed up the die size with the extra core functionality to support the features, required more SPs, etc. to be able to run those features at a useable performance level and so on... then you come to the software development to support all these new features... all in all if they had developed cards based on the original specification and given them enough hardware to produce a useable performance level we'd be looking at a big increase in costs to the end user...
Implementing DX10.1 on 65 or 55nm processes is another ball game.