Re: Simultaneous Switched Matrix (SSM) speed versus Xeon
Complex question.
1) SSM doesn't compete against the Xeon. In fact, it could eventually support it. It does compete with Intel's shared-bus architectures (450NX chip set for 4-way and Intel's Corollary unit's Profusion product for the 8-way market). With a bus-oriented approach, there would be a need for arbitration and the greater the load, the greater the arbitration. If I understand it correctly, this leads to a scalability question, which Profusion will address by use of a second bus.
2) The network approach (SSM) obviates the need for arbitration due to the fundamental difference in architecture. Scalability is constant.
3) While Poseidon's technology would be appropriate for all MPUs, it would appear to make more business sense to align it with the non-Intel part in the beginning. After all, AMD needs a lever into the server market, while Intel has an investment to protect. Interesting convergence of needs shared by AMD and Poseidon.
So how about the market? For next year, 2-way systems will probably be about 2MM units (4MM MPUs), while 4/8-way systems might be 1/4 of that (or .5MM servers X maybe an average of 5.5 MPUs/server or 2.75MM MPUs).
This is why I selected the K-7 as my semiconductor product of the year back in January, citing the Poseidon technology as a driver. Starting from nowhere last year and only getting into the server market later this year, AMD could get a meaningful percentage of the market in 2000. Make your own assumptions (e.g., AMD gets 750,000 K-7s for the server market packaged in a slot configuration so they sell for $500 each, while cost per part averages $200; AMD would have an incremental $300*750,000=$225MM in gross margin, just to create an example) and you can get an idea about the potential effect on AMD earnings. The execution issue doesn't go away, obviously, and the risk in the stock remains, but think of all those servers with a $4,000-$500)*8=$28,000 difference in MPU bill of materials between a Xeon/2MB approach and a K-7 approach if the K-7 goes for $500.
Pretty interesting, don't you think? - Tad LaFountain |
