I don't know why you need any more evidence of the value of performance leadership than the fact these companies pour millions of dollars into attaining it. If there was no value why would they do it?
It's all about establishing and promoting a brand. Nvidia and ATI are brands. G80 and R600 are brands. It's those same industry insiders and enthusiasts that are best equipped to acknowledge the quality of mainstream products in isolation. But a lot of regular folk buy brands. ATI didn't produce a 700M transistor part because they thought performance leadership was unimportant. And that is the only context in which the situation can be evaluated IMO.
ATI no longer exists, the ballgame may have changed, at least a year or so from now, sufficiently far along the design pipeline to make changes in direction.
That said, it is obvious that AMD and previously ATI feels compelled to compete at the high end. It is the nature (still) of technology enthusiast reporting to focus almost exclusively on the high end, and having top performers (helped by driver optimisations) ensures that your name is glued to the top of every benchmark chart around.
But the value of this is hard to quantify. ATI was murdered in the marketplace in spite of having the 19X0XTXs, arguably the fastest parts at the time, and lost marketshare very rapidly both in mobile and stationary applications. AMD, in spite of their rebranding of Opteron->Gaming dual socket systems, have still lost a lot of marketshare to Intel. Not only are the effects of brand building difficult to assess, brand building through having a high performance offering is a special case. How useful is it
really? Suffice to say, it doesn't seem as useful as actually being competitive in the market where a consumer is interested in buying, which is actually a pretty healthy sign.
And it doesn't just come down to price/performance, ATI used to compete successfully in the mobile space based not only on performance, but also on thermal engineering which simplified the design of the whole computer, enclosure and cooling, and gave the consumer quieter computers with longer battery lives. Driver quality is likewise important to consumer satisfaction, and so on.
Obviously the HD2900XT is a disaster. Late, underperforming and overheating. But even if it had matched or exceeded the 8800GTX, would that have sufficed to make it a success? Would it have made sense? Would anyone, except for a handful, give a damn regardless of the benchmark charts? This is where the marketshare data vs price level, and the Stream survey, now over half a million replies, tell a pretty definitive story - No, it wouldn't really have mattered much. The 8800, uncontested champion in performance
and features for 9 months (!), has achieved a whopping 1% penetration among online gamers. (And at this low level, industry insiders are likely to significantly affect the data, not to mention that the WoW/Sims part of the market is likely even less interested in the top-end offerings)
The high end monster card path is, IMO, a dead end in terms of technology evolution.
Regardless of technology though, what is more important is that the market rejects them. People just don't buy, in spite of advertising and tech reporting hype. So maybe it would make sense for the manufacturing and reporting industries to start asking themselves what people actually desire and spend their efforts there. It's not likely to be GPGPU and Crossfire....
The market data really should be stickied - so much discussion here is based on ideas about the market that are just plain wrong.
), this is a very bad news from user aspect, they miss unreal3 and crysis without any performance GPU, most of the users will upgrade there computers for this 2 game this year.
) then your GPU performance is exclusively limited by how many mm² you are willing to dedicate to it and your raw processing performance per mm² (at your target die size, thus taking scalability into consideration).
