Vol 16, Issue 12

March 25, 2002

Free Advice for AMD and Intel

By Kevin Krewell

Free advice is usually worth what you pay for it. It often comes from people without full knowledge of a situation. It is suspect because if it is so important, so astute, why is it free? Well, at times, we all feel compelled to butt in, and I am no exception. It sometimes takes an outsider to look at a situation and make a common-sense suggestion to clarify the issues involved. In the name of journalistic fairness, I have balanced my suggestions by offering advice to both AMD and Intel in the same editorial.

I believe someone has to point out when the emperor has no clothes, and in this case, the missing clothing is customer demand for AMD's SledgeHammer processor. To be blunt, I doubt there is a significant market for AMD's SledgeHammer, and I think AMD is placing engineering resources on the wrong opportunity. Before I explain why I have come to this conclusion, let me say that I consider AMD's Hammer family an excellent engineering solution. I also see AMD's evolutionary 64-bit migration plan for the x86 instruction set as a good conservative strategy. I believe the ClawHammer processor will gain acceptance as a high-performance processor for desktops, notebooks, workstations, and one- and two-processor servers. AMD probably senses that Intel's Xeon processor is vulnerable because its very deep pipeline and small L1 caches are not efficient on server applications. I have a problem, though, with AMD's chances for SledgeHammer in the low-volume, high-margin, high-barriers-to-entry, and ultraconservative server market for four-way and above processor configurations.

AMD's SledgeHammer is designed for multiprocessing systems that use from 4 to 16 processors. With coherent HyperTransport links between processors, AMD can gluelessly connect arrays of SledgeHammer processors, not unlike the capability of Compaq's forthcoming Alpha EV-7 processor. AMD has stated that SledgeHammer will allow it to penetrate the server market by offering a complete solution, from one processor to many, as Intel does. Although that claim is technically correct, what major OEM would actually design and build a server using SledgeHammer? And, perhaps more important, are any AMD customers asking AMD to build SledgeHammer?

A quick review of the top five server OEMs, which (according to Dataquest) have 70% of the server market, does not provide a promising candidate. Compaq has already committed to Itanium, having killed the future Alpha EV-8. Compaq may shortly merge with HP, and HP co-developed the Itanium architecture, making it very unlikely that HP, or a merged HP-Compaq, would choose SledgeHammer. IBM already has the excellent Power4 and has support for Itanium, giving it no incentive to add another 64-bit architecture. Dell has yet to field any non-Intel solution. Sun is committed to the UltraSPARC architecture but will sell Linux boxes using x86 processors. To avoid competing with Sun's own high-end servers, Sun's Linux boxes will likely be limited to one- and two-processor solutions and be very cost sensitive. Those boxes offer a good opportunity for Athlon MP and ClawHammer. SledgeHammer systems larger than dual-processor systems would be too threatening to the UltraSPARC interests at Sun.

This situation leaves AMD with high-end, tier-two server candidates like Cray, Fujitsu-Siemens, SGI, and Unisys. These smaller vendors would have to spend millions on system design and validation for a processor that has no market share. If AMD is counting on Itanium's eventually failing to be competitive, as its critics have maintained, Intel may still have the rumored Plan B for Xeon—Yamhill. Eventually, 64-bit computing will be essentially free, from a silicon standpoint, and Intel will add extended addressing to the x86 architecture.

A much better return on AMD's engineering investment would be provided by focusing on a new mobile-specific processor. When AMD went from zero market share to roughly 40% of U.S. retail notebooks in 2001, it showed it could swiftly grow market share, given a competitive product. Furthermore, AMD might also be able to leverage the expertise in low-power, high-performance processor design from its recent Alchemy acquisition.

Intel's Banias processor will likely change the market for notebook processors in 2003, when clock speed and core voltage control schemes will no longer be sufficient for a mobile processor to be state of the art. Intel is poised to raise the bar with lower-leakage power management, dual V_T semiconductor process technology, and specialized chip-set support with integrated wireless LAN capability. We also expect Intel to have microarchitecture-level power management controls.

The notebook market is growing faster than the market for either desktop PCs or servers. Mobile processors may not have the very high prices of server processors, but mobile is a market in which AMD has had a successful track record and could provide a greater number of its customers with a product they want and will actually use.

The Name Game at Intel

In my last editorial, I railed against marchitecture, but for the moment I would like to embrace one of my favorite marketing games: naming processors. I think this is a timely discussion, because Intel is about to make some major changes in its processor lineup, and I think there's still time to help it make the right choice.

In mid-2002, Intel will take the Celeron product from the P6 generation to the P7 generation (which Intel calls the NetBurst architecture, refusing to get into the generation-number game with AMD). The NetBurst version of Celeron is a new-generation product with a new socket, new instructions, different performance characteristics, and several other innovations. When the change from Socket 370 to Socket 478 versions of Celeron occurs, it should not be done without a clear marking that indicates to OEMs, VARs, and consumers that the "new" Celeron is significantly different from the old one. Intel should add a modifier to Celeron brand name to clearly indicate the distinction. The modifier could (and should) be as simple as a "2" (as it is the second microarchitecture generation for Celeron) or a "4" (a reference to the similarity to Pentium 4) postfix. I cannot sufficiently stress how important I think this clarity will be for Intel's customers.

In addition, in 2002, Intel launches its second 64-bit microprocessor—McKinley. If ever a product needed an image booster, it's Itanium. The obvious name change would be Itanium-2 or Itanium II. Some wags have suggested Anadium, derived from the next element in the periodic table after Titanium. (If Itanium is titanium minus the T, then Anadium is vanadium minus the V.) I'll leave it to the highly paid name consultants to decide whether it should have Roman numerals or Arabic and be hyphenated or not—although I'm fond of the retro Roman numeral version (sans hyphen).

If AMD or Intel is reluctant to follow this advice, it does come with a full money-back guarantee.