Ta, I wasn't insinuating that Titans memory would be any cheaper just FYI as that would be silly. Just wasn't really convinced that the price difference (assuming the 450+VAT is semi accurate) would be large enough to account for the £900 price tag.
Not even slightly, Titan is a pee take, the 780 is the same chip(or at least afaik) which has a more sensible(but still not sensible at all

) price. Ultimately it's very hard to know the yields.
Roughly speaking you have 300mm wafers, so pi*r^2 gives you a wafer area of circa 70500mm^2, divide that by chip size and you get a bit under 130 chips per wafer for GK110, and around 160 for Hawaii XT or about 24% more chips. That assumes square shapes fit perfectly into a circle which obviously they don't. The smaller the chip the less wasted space so a Gk110 wafer could easily be (when you think the outside edge is largest and could effect almost every chip on the outside edge) anything from 5-15 less maybe. Titan is expensive to make, so is Hawaii, but it's not "that" expensive theoretically. What you get into is R&D costs, you might be making a chip that costs $250 with bad yields, but if you're trying to recoup 500mil R&D costs, you add a lot to that price before you even make a profit, which adds more again.
So it's very hard to judge where break even point is. I would think 2 years on from 28nm even Titan will be yielding well enough to be sold relatively cheaply and still make a profit. When you're talking about selling maybe only 5-10 parts from a wafer that are fully working as $2k professional parts, and maybe getting 20-60 more you can sell as various levels of cut down parts as Titan's/780gtx, I'd be very surprised if Titan's couldn't make an effective profit being sold at £500 and 780gtx's way less than that.
I really think if the market just turned it's nose up at Titan's prices they would have crashed it quite happily to £500-600 and sold with a profit. Nvidia just priced it high to see if it would get any sales, nothing stopping them dropping the price by £300 over a few months if it didn't sell, AMD had nothing planned for a while, they had nothing to lose and a decent amount of cash to gain.
Ultimately I don't know if AMD will try the same thing, and if then do and people still buy it, we can look forward to £500 + launch cards every gen from now on at least.
A sensible guess is for any given price AMD can make a profit on the R290x(which could be no where near the sale price) then Titan chips should cost a minimum I would think 30% more for the chip alone(with similarly good yields), probably closer to 40% more if Nvidia's yields aren't good(historically they've been meh for a few gens there) then the pcb/memory costs.... if that was more than $50 extra I'd be surprised.
SO if AMD could sell a R290x at £400 for decent profit, a titan should really be at most £600.