FErmi is and was a HUGE failure, today, yesterday a year ago, 5 years from now.
Its really that simple, marketshare, AMD has complete control of dx11 marketshare, Nvidia still doesn't have proper low end parts because of the modularity that means they just can't put out a 16 shader card like previous gens thats tiny and costs nothing and sells upwards of 10mil to OEM's.
Fermi, its power issues, and Nvidia's general behaviour are almost certainly screwing them out of a chance to be in a single next generation console. So imagine whats going to happen when not a single console, or console to PC port, or multiplatform release isn't "nvidia optimised".
I've also said, for OVER a year without changing my tune that Fermi is a massive failure for NVIDIA not the end user.
Frankly give me a 600W uber card witha triple slot cooler, I'll buy an 1200W psu and be very happy, the cost of a mammoth single core that used that much power is something Nvidia/AMD can't afford to make.
Fermi is incredibly NOT profitable, at any price point except the 580GTX which while a few people do have them, and a few in terms of all graphics sales is still 1000's and these forums give an entirely skewed idea of whats in a "normal" persons system, its not selling in vast quantities AT ALL. Its huge, and it sucks power, and OEM's don't really trust them, and they've lost market share, and they can't offer a platform, and they are losing sales and money over it.
Nvidia make PLENTY of money on other area's, but desktop gpu's Fermi is an entire failure, Tegra 2, still to be seen really, Nvidia finally got something right, first to market with a dual core arm is a pretty big feat, though is it best, who knows. The question is, does best matter when you're first to market and established, ask Nvidia the answer and mention DX11 and desktop Fermi architecture sales.
Faster doesn't mean better, or a success, if AMD turned around and sold 6970's at £30 they'd have the most successful card of all time, and lose hundreds of millions of dollars, great card at a great price, success, no. If they could ONLY sell the 6970 at £30 it would be an entire failure even if end users love them.
THe 6950 makes more profit than a 570gtx, sells more, isn't far off performance wise and with a free and easy flash is basically as fast/faster than it.
This is the problem, success and failure are quantifiable, is the 570gtx a success with users, yes, is it a business success driving forwards massive sales and market share gains vs their opposition, not in the slightest.
The only reason the 580gtx is faster than AMD's current top card, oh wait, its not, its still 40% behind the 5970........
Seriously the only reason the 580gtx is faster than the 6970 is the 580gtx IS the design Nvidia planned and tried to make to be released late 2009, that had been worked on for a couple of years at that point, and took 4 respins(or 3 I've lost count now) to get working right. The 6970 is a MUCH newer design, started MUCH later entirely designed for 32nm, that had to be brought to a higher process design and have lots of bits cut off to fit, they did this in a much shorter space of time, on the wrong process, improved performance, added features, improved flexibility, improved gpgpu features, added features that will help push them in a professional market, drastically increased crossfire scaling.
In the midrange they got almost the same performance out of a 255mm2 core, thats HUGE, business wise thats absolutely immense for them, thats a hugely profitable and very fast gaming core. Huge sales, huge profit, huge marketshare, huge availability, huge success.