SOI, AMD, Fabs, and an impending cash crunch?
<< But what if SOI works out the way IBM, Motorola, AMD, and a bunch of other companies say it will? AMD could be in a position to swap Austin for Gresham next year.>>
I'm a heavy Mac users, meaning I have G3 and G4 processors in my laptops and desktop machines. These processors are from Motorola and (I believe one of them) from IBM. Certainly IBM and Motorola are both part of the PowerPC Consortium. And yet the "Megahurtz War" is currently going badly for Apple: WHERE IS SOI FOR THE PPC? If SOI is so much faster, and if Motorola and IBM have a manufacturable, economical process for SOI, why has it not been applied for actual chips being sold to actual customers?
If Motorola and IBM are so danged far ahead in SOI, where are they for the very chips they say will beat Intel? (Actually, IBM has bet rather heavily on the Intel architecture, in servers, PCs, and even workstations. Moto seems to be lukewarm about the PPC. Whither the PPC?)
<< By Q3 of 2003, AMD could be producing 20 million X86-64 SOI Hammers per quarter. If they clock as high as P4, have better IPC, (and 64 bits, good for marketing at least) and use less power than P4, AMD might be the one with $150 ASPs. Meanwhile Intel is left trying to cover their $7 Billion per quarter costs while selling 30 million P4s at an ASP of $75. Hopefully "other" would be doing better by then, but losses could still easily exceed $2 Billion per quarter.>>
There's an old investment joke about a customer being shown the yachts of J.P Morgan and the other brokers. "But where are the _customers'_ yachts?" he asked.
Were I an AMD advocate, I'd be asking "Where are the new fabs?"
(I know you, Dan, have claimed that AMD's reliance on basically just their single mega-fab, Dresden, implies that they must have superior yields to Intel and that the reason Intel has built several new fabs in recent years, with more on the drawing boards, is that Intel clearly has abysmal yields and must spend tens of billions to stay in place with AMD. A fanciful theory, but amusing.)
Without massive new expenditures for fabs, how can AMD churn out these chips in the way you think they can? I see AMD just not having enough cash to build these new fabs, and not being willing to go deeper in debt to do so. Maybe they'll buy the Gresham plant, or other struggling plants, and try to make a go of them.
(The history of buying older plants and using them for cutting-edge uses has not been bright. Usually older plants get the older products. Lots of reasons for this.)
<< The whole industry, except for Intel, went for DDR instead of RDRAM - and Intel was wrong. Now the whole industry, except for Intel, is now going for SOI - what if Intel is wrong again? >>
Intel had an active SOI program in 1974. Colleagues of mine worked on it (Dave Wanlass, others). We looked at it several times later, too. I assume they have an active program now as well.
It's not some magic bullet. Just another technology.
To assume that AMD, without having built any new fabs in the past few years, and with (apparently) none on the drawing boards right now, is going to deploy SOI as some magical Intel killer is just plain wishful thinking.
I try not to be an AMD basher here. In some ways there presence in the Intel architecture market has helped Intel (second source, price competition, helped give customers alternative designs, etc.). The future IS the Intel architecture...the competitors have dropped out. If AMD has 10-15% of this market, fine with me.
However, I doubt they can even hold on to 10-15% of the market, given the trends. Intel is proliferating designs, has half a dozen fabs making IA chips, and will be building more plants soon. AMD was successful with Dresden--give them credit for that--and with Athlon--ditto--but how about the FUTURE?
It sure looks to me that AMD is running desperately low on cash...certainly it does not have the billions in the bank that Intel has. A cash crunch could crimp chips.
--Tim May |
