Originally posted by: dreddfunk
Cold - you and I pretty much agree on this issue. I'm just a little bummed that the performance segment is always the last to be addressed. Sure, you eventually get parts like the 7900gs or x1950pro, but it's usually long after the high-end and mid-ranged cards are around. I do understand why they're doing it--because performance leadership and mainstream sales are more important--but it still stinks for those of us who want good performance (read: not HD2600XT/8600GTS performance) without paying exorbitant prices. I'd really like to see a 256bit, 512MB, 8900GS with 96 shaders.
I agree with what you said, the enthusiast parts need to be released first and be the shining beacon of light to the rest of the line, while the mainstream needs to be released not too long after to take advantage of the hype and generate profits at good production cost.
The performance mainstream segment is simply too difficult to sustain at the beginning of cycle, due to it's need for good performance and good price, in essence, no concessions can be made, or the leeway provided is simply not enough.
Hence you only see good performance mainstream products at good cost, later in the cycle, Geforce 6800 (NV42 version), Radeon X800 XL (R430), Geforce 7900 GS/GT (G71), Radeon X1950 Pro (RV570).
If you have to do it early in the cycle, you usually make the sacrifice of production cost in exchange for performance.
Radeon 9500 Pro (R300), Geforce 6800 (NV40), Radeon X800 Pro (R420), Radeon X1800 XT 256 (R520), Geforce 4 Ti4200 (NV25).
Or vice versa with production cost being emphasized over performance.
Radeon HD 2600 XT (RV630), Geforce 8600 GTS (G84), Radeon X1600 XT (RV530).
Given how large the gap is now, you probably will need 1 full node process later before you can fabricate an effective performance-mainstream without making concessions on either side.
In prior generations the gap was smaller so only 1 half node was required to make an effective performance-mainstream.
Unfortunately there isn't much you can do about this if Nvidia and ATI want to keep sustaining the performance leaps on the enthusiast end, process technology just can't keep up with the rate they wish to plow ahead.