smooth,
Intel has shown Intel data on successive generations of yield graphs that say the rate of yield improvement with each generation increases through to 65nm.
Those graphs, if you noticed, don't label the Y axis, with anything concrete, but instead, with something called Mature yield, which is open to definition (and redefinition) for each successive curve, because the author of the graph (Intel or AMD) can have a different standard of what constitutes a mature yield at each node.
Another thing is the whole idea of yield. The general definition is #of good die per wafer. But a "good die" that can perform the work without errors can:
a) not be able to achieve the desired clock speed
b) leaks too much power.
Think of the challenge Prescott generation faces: Power consumption is too high, and bin splits are below target.
So Intel can have a lot of "good die" that unfortunately fall into the bad quadrant of bin splits and power consumption, and just cannot be sold, ot they can be sold as Celeron, but Intel already has too many of those.
I am not saying that this is the case, but it is just a possible explanation of good yield of working parts, but not good enough yield of the desirable parts.
Now, with DC Smithfield, things are so much tougher. The bar of how low the power consumption needs to be for the 2 die to add up to less than the limit is much tougher.
My WAG is that yield of saleable parts of Smithfield is atrocious, limiting most to the lowest bin or below (unseleable). Then, on the other hand, the performance of Smithfield is very bad compared to competition, not justifying anything other than a deeply discounted price.
I think this could potentially turn around somewhat with 65nm Pressler, with improvements in average power consumption. Further physically detaching the die will improve the yield, and thirdly, rumored introduction of even lower bin than the current bottom one will turn the former dumpster material into a seleable part.
Given that Intel has maintained their lead throughout these generations and that the 45nm lead seems ensured as demonstrated today, what, except for a good quarter from AMD has *changed* that AMD is going to hold more than a historical 20% of market share in light of Intel producing an entirely new core in the beginning of H2 that threatens to take the performance crown, take back any market share advantage AMD has and leaving the only advantage that AMD has is an "elegant" architecture that the 'droids have fallen in love with?
That is certainly a possibility, but the big hurdle to overcome, for Intel will be Q2. AMD will be capacity unconstrained. Q2 is the slowest quarter, and AMD hs a ton of capacity coming online. Anybody with any dissatisfaction with Intel will be free to jump the ship (Intel will be too afraid to retaliate because of the law suits). That can translate into a "bad" quarter for Intel (earning some fraction of a billion less) but most importantly, bad PC from lost market share.
After that, with NGA, things can turn around for Intel. Who knows.
Joe
AI
