"AMD does improve processors. So does Intel."
First, unless things have changed, AMD can manufacture fairly large die economically, the original K5 was 251mm^2 and the original K7 was 187mm^2. What AMD hasn't done is made a L2 cache larger than 256k. Can they? Now that's a question...
Pretty much a given that the two companies improve their processors. But, unless there is a huge surprise somewhere, we know what the next few years will be like. Next year will have at least two flavors of the Hammers, and Prescott. I suspect that Prescott will have more resources to support Hyperthreading better, a very fast FSB and 1 Meg of L2. By the time it is out, the Hammers will either have started a transition to 90nm, or will be getting ready to. I expect AMD to do two things at 90nm, add DDRII capability and increase the caches. However, even without an increase in cache, a straight ClawHammer shrink to 65mm^2 with a 512k L2 cache and DDRII should be a potent competitor to Prescott unless ClawHammer's clock rate cannot exceed 3.5GHz at 90nm. Since SledgeHammer with it's 1 Meg. of L2 will be in the 150mm^2 range at 130nm, it should be in the 100mm^2 range at 90nm, assuming it scales down by the same factor as ClawHammer is supposed to. So AMD has to decide whether to go for a 100mm^2 die, add more cache, add a second processor core, or all three and have a new level of performance. I suspect they will go with the latter and have the desktop ClawHammers fabbed out, keeping the Sledge variants at Dresden. But that's my guess.
Ok, that accounts for up to 2004 and early 2005. 2005 should be the transition to 65nm for both companies and AMD's transition to 300mm. If AMD still has capacity problems after that, well, that should be very good for AMD unless 450mm^2 chips are needed... |
