Elmer: So trying to divine Intel's yields is hopeless while AMD's are possible to estimate fairly well.
We certainly agree that estimating Intel's yields is hopeless - not only do they have a more complex product / fab / process distribution, they are also not as informative about actual shipments (so we don't even have a good # for total units, much less for each flavor of x86 processors).
With regards to AMD, however, I still don't think it is possible to estimate yields "fairly well"… I know there is SOI and .13µm bulk development going on at Fab30 in Dresden. I have zero idea, however, what the "wafer cost" is. There's also new stepping development and some Hammer development…
I also cannot come up with a reasonable estimate of the average equipment downtime in a Fab such as AMD's Fab30. And then there's the "equipment downtime" factor, which I've been told is non-trivial (however, I assume these things do not vary a whole lot from fab to fab, so others with good fab-knowledge could presumably come up with a reasonable estimate - yourself perhaps?).
I wouldn't be surprised, however, if AMD's TBird yields were around 70-75%, with Intel's Coppermine yields coming in something like 5-10 percentage points higher (standard process, not the ultra-tweaked notched gate thing, of course).
-fyo |
