Bill, Re: "There will indeed be some kind of contest between the downwardly compatible Hammer and the IA64. It is not certains that Intel will win this race because people are reactionary and with a choice between the two, one of which runs the old SW and one of which runs the new(small number of titles) how will they jump?"
I don't believe you are looking at this in the right way. Hammer is set to launch for UP and DP servers in the first half of 2003. Ask yourself what will be the installed base at that point.
Since McKinley is launching in the middle of 2002, I believe it will almost be starting at zero, just like Hammer. Sure there will be some early adopters, but I don't expect OEMs to start offering robust McKinley solutions until the end of 2002, or perhaps even by the time Hammer launches. Therefore, both will have an opportunity to gain in market share, if they can exceed the benefits offered by other solutions.
So what will those solutions be?
Right now, Intel controls about 90% of the UP and DP server market, 60%+ of the 4-way mid-range market, and about 25% of the >4-way enterprise market. My guess is that McKinley will be targeted at growing in the latter segment. I doubt that Sledgehammer will have enough infrastructure to even attempt this space right out the door, so at the best, it might be competing for the 4-way mid-range space, at least to start out. Clawhammer will be competing for the UP/DP volume space.
Currently, Since Intel is the majority leader in these spaces, AMD's best chance is to offer something significantly better than the installed base. Remember, that most businesses won't be switching from their current infrastructure without a good reason, so Hammer must be able to offer a much better value than what they are currently using.
Most probably, Hammer will be competing against Xeon, not Itanium (Itanium meaning McKinley, Madison, Deefield, or whatever Intel will offer at the time). The Xeon infrastructure will soon be moving to Netburst micro-architecture. Starting early next year, Intel will offer their large cache version version of Foster with Hyperthreading enabled, called Foster MP. Foster MP will work with the Grand Champion chipset for 4-way servers, and by the end of next year, will go through the .13u shrink of Gallatin. Gallatin will offer 2MB of L3 cache, as well as all the features of Foster, and probably have clock speeds higher than Hammer.
Sledgehammer will have 1MB of L2 cache, no SMT, and will benefit from the micro-architectural improvements of AMD's K8, as well as the high-speed Hypertransport interconnect. I've seen AMD's implementation of 4-way designs, and it certainly isn't as efficient as it could be. Quite possibly, it may not scale as well as the large cache Xeons. But even if it does, will it scale enough to offer a large enough advantage, such that businesses will discard their Xeon infrastructure in favor of Hammer?
In terms of UP and DP markets, Intel will have already had Prestonia for a year before Hammer launches in the server space, which makes Clawhammer a competing technology to the small cache Prestonia Xeon infrastructure. Like Foster MP, Prestonia will have Hyperthreading support, and will likely reach the high frequencies of Northwood. It will also have 512KB of L2, like Clawhammer. Again, the question appears - will Hammer have so much of an advantage that businesses will abandon their Xeon infrastructure in favor of Hammer?
Finally, where does this leave Itanium? That's the key question, since in the UP, DP, and 4-way spaces, Itanium would also likely be competing with Xeon. Therefore, I think it will take some time before we see Itanium get to be very large volume.
On the other hand, Itanium will be an excellent solution for the >4-way space, since Alpha and PA-RISC will slowly be moving aside as Compaq and HP transition their lines to Itanium. Further, Itanium will likely make its way into the 4-way space by competing further with Sun. Performance wise, Itanium still offers more than UltraSparc, and disparity between the two is only likely to increase over time. Intel will have a hard time convincing businesses to abandon their stable Sun infrastructure, but Intel does have the horizontal model working in their favor, and hopefully a much smaller TCO will convince some businesses to make way. Additionally, some of the Xeon market will transition to Itanium, but only on the whim of OEMs like Dell and IBM. It is these who can make or break Itanium.
But as for Hammer, without any first-tier support, how will it gain any ground? It's not going to be the ubiquitous choice that you think, at least not right away.
wbmw |
