He tells us that, "On the server side, there are two schools of thought. One says Intel is going to be the future as far as we can see (a few years down the road). And then there are others like me that say Intel right now, on the low end, has done a good job of characterizing workloads." I for one do not accept the idea that Intel's success on the desktop is necessarily transferable into equal success in the data center. Intel might well succeed anyway; they certainly have fine engineers and a track record of being able to manufacture their designs. But I don't believe the volumes they achieve at the low end are going to be all that big an advantage in the 'big box' world. They didn't seem to think so either, and so went ahead with a software-incompatible Itanium. In a vacuum that might have worked, but AMD has thrown them a curve ball in the form of a 64-bit processor that will natively execute the huge base of 32-bit code. AMD's design could even be somewhat inferior come the actual future, but it is much better at solving the real-world problem of how we get there from here. IT managers don't get to leap into the future in one hop; they're going to be stuck with some chunks of 32-bit code they have to support during the transition. AMD just plain has a better story than Intel on how to do that. In addition, because of the Itanium's slow sales, Intel has nothing like the volume-driven manufacuturing cost advantage they have over everybody else in the "P4" space. Before too long, AMD's Opteron could be shipping in higher volumes than Itanium. IBM's Power series and SUN's SPARCs probably already are. So Intel is no slam dunk in the server market. Our protagonist unfortunately does not go on to tell us what he thinks the 'other' alternative is. It could come from either direction, and blindside Intel badly. Whatever the hell those little chips are that are going into all those cell phones, they are going to get dirt-cheap in the volumes they are being made. In transaction-oriented environments like web services, a bizillion little chips is every bit as good as one Big Kahuna processor. So the Itaniums, the SPARCs, and the Power chips could all find themselves getting whacked by Redundant Arrays of Inexpensive CPUs... the RAID idea applied to server boxes, with cell phone chips used as the CPU's because they're so damned cheap. There are still lots of jobs that nobody really knows how to break up all that easily, which is why IBM and Sun are still able to make money peddling "big iron." In those boxes, the CPU chips per se are not the determinant of performance that they are in simpler boxes, so even low-volume proprietary chips like SPARCs can find a home if the rest if the box design is appropriate. For these reasons, I don't see how Intel has any particular advantage at the high end, and at the transaction level where today they are moving lots of cheap server boxes for Dell, HP, IBM, etc., they might get surprised by something out of left field... like the Second Coming of Motorola, or one of the other cell phone guys. Sun would be a good candidate to come up with a "One thousand cell phone chips in a box" product, because that kind of architecture is right up their alley. IBM probably has one in the back room, too. It's an obvious idea. It's basically the same idea Sequent had, back when putting 64 Pentiums in a box was a big challenge. |
I was disappointed that the guy either didn't finish his thought concerning possible Intel domination of the server market, or the article got chopped up in such a way that it was lost.