Couldn't it be simply be to increase margins in these potentially high volume parts? Higher performing chips are already profitable, and will be lower volume.
I argued the same thing when a roadmap 6 months ago revealed that AMD would be making Paris chips on 130 nm all through 2005 (IIRC). I said, why - they will waste more of valuable production capacity to build them on 130nm than if they were built on 90nm?
Some others chimed in that Intel liked to continue building their budget chips on older process technology because they could use "fully depreciated factories and equipment" to do so. Therefore the cost of production was low even on the prior technology.
But that argument holds no water for AMD. At AMD, additional wafer starts for building Semprons on 130nm requires more of the factory production, and therefore it DOES make sense to build them on 90nm, BUT ONLY IF 90NM YIELDS ARE DECENT.
{see last paragraph for why AMD originally planned NOT to use 90nm for Sempron until late 2005}
But I still see other evidence that 90nm is not producing faster-clocking parts yet -- 1) the fact that the fastest chip released so far is a 3500+, not a 3800+, 2) the fact that they chose to produce notebook chips first rather than an Opteron 252.
So, I conclude that 90nm yields are good, even better than expected so that AMD can use 90nm to reduce the fraction of the fab needed to make Sempron chips. But I also conclude that 90nm is not yet yielding higher-than-130-binning parts.
It really took Intel almost 6 months to get 90nm binsplits better than 130nm and, in fact, they might have just broadened the binsplit range to get a few parts faster than 3.4 GHz. If AMD can get a 90nm 4000+ part by the end of the year (3.5 months), it will be a major victory.
I never explained why AMD originally planned to continue 130nm castrated-K8 production for so long. Two reasons I think. First, Sempron demand exploded because of the China thing. But before they knew that, the demand wasn't going to be very great, so they wouldn't have to use 1/3 of the fab to fulfill it. So 130nm was reasonable, pre-China; it's not, post-china. Second reason is that AMD was planning for the worst case where they would be cash-constrained w.r.t. capital spending to convert to 90nm. But now AMD has enough free cash flow to convert to 90nm sooner rather than later.
Petz |
