The 64-bit battle
economist.com
Is Intel's flagship next-generation chip already doomed?
THE computer industry loves a good bunfight, so everyone is enjoying the current battle between Intel, the world's largest chip-maker, and Advanced Micro Devices (AMD), its smaller rival. Intel is currently on the defensive. This week Hewlett-Packard (HP), a computer maker and a close ally of Intel, said it would use AMD's Opteron processor chip in a new line of powerful server computers. The news came just days after Intel grudgingly unveiled a new version of its Xeon chip, which mimics features of the Opteron. AMD, it seems, has Intel on the run.
At issue is the best way to move the industry from 32-bit to 64-bit chips. Except at the very high end and in some specialist niches, almost all computers are currently powered by 32-bit chips. As their name suggests, 64-bit chips can process data faster, in chunks twice as big. They also make possible computers with greater memory, since 32-bit chips are limited to four gigabytes (roughly, four billion bytes) of random-access memory. Just as the industry previously moved from 8-bit to 16-bit to 32-bit computers, the transition to 64-bit is inevitable in the long term. The question is how to manage it.
With Opteron, AMD's approach was to take an Intel-compatible 32-bit chip, and add special 64-bit extensions to it. The resulting chip is able to support more than four gigabytes of memory, and can run new, faster 64-bit software. Crucially, it can also run existing 32-bit software in the usual way. Intel's approach, in contrast, was to start with a clean sheet and design an entirely new 64-bit chip, called Itanium. The project is in fact a joint venture with HP, which began work on the chip in 1988. Since its launch in 2001, years late, Itanium has sold slowly: a mere 100,000 chips were sold last year. Itanium systems are powerful, but they are expensive and cannot run existing 32-bit software efficiently. To get the full benefit from Itanium, software must be extensively rejigged.
Fred Weber, technology chief at AMD, likens Itanium to Esperanto. Nobody wants to learn a new language, he says, no matter how elegant it is. Hence the appeal of AMD's chip, which offers cheap 64-bit power without compromising the performance of existing 32-bit software. HP's announcement that it will use Opteron chips, and Intel's announcement of an Opteron-like Xeon chip with 64-bit extensions, therefore appear to be very bad news for Itanium, and a vindication for AMD.
Intel denies that Itanium is doomed, however, with some justification. Itanium is intended to compete with other 64-bit chips (such as those made by Sun and IBM) at the very top end of the market, says Lisa Graff of Intel. That is a totally separate market from the low-end servers powered by Xeon and Opteron chips. Around 85% of servers sold cost less than $6,000. But the 15% that cost more account for 50% of server sales by revenue. Itanium is aimed at this low-volume, high-end market. HP, for its part, says that it remains committed to Itanium in its high-end systems.
Another factor in Itanium's favour is that its performance is expected to improve dramatically over the next few years, notes Dean McCarron of Mercury Research, a market-research firm based in Scottsdale, Arizona. By 2007, Intel expects Itanium and Xeon chips to cost about the same, but the Itanium chip will be twice as fast. In other words, Itanium will start at the high end, and will slowly move down towards the mass market. The trouble is that Intel's previous predictions about Itanium have been wide of the mark.
The big question now is whether Intel can maintain a clear distinction between its Xeon and Itanium product lines until Itanium's price comes down. Just because the Opteron and the new Xeon chips are 64-bit, Intel argues, does not mean that they compete with Itanium. There is more to a high-end chip than just 64-bit support: such chips also need special reliability and availability features that Opteron and Xeon currently lack. The danger, notes Peter Glaskowsky of Microprocessor Report, an industry journal, is that AMD will add such features to Opteron, forcing Intel to respond in kind, cannibalising features from Itanium and adding them to Xeon. This would condemn Itanium to a high-end niche, or worse, render it redundant.
The fate of Itanium, in short, is still unclear and depends as much on the actions of AMD, HP and others as it does on those of Intel. “This is a chess game,” says Greg Papadopoulos, Sun's technology chief. But at least Intel now has a 64-bit insurance policy: if Itanium fails to move down into the middle market, Intel can push its Xeon chips upmarket instead. |
