AMD must focus on cost to beat Intel – report
90nm or bust
AMD must rapidly move to a 90nm fabrication process and shift quickly from its ClawHammer Athlon 64 to the 'Paris' core if it's to maintain or even grow its share of the x86 processor market, researcher In-Stat/MDR has warned.
It will also need to recruit more tier-one vendors alongside IBM if it's to make Opteron as successful for AMD as the chip needs to be, the market watcher has said in a new report on the x86 chip market.
Essentially, AMD's biggest problem is cost. In-Stat/MDR's line is that to continue to compete effectively with Intel, which it does in the desktop, mobile, server and value markets, AMD has to control its costs.
That means moving quickly to 90nm, in an attempt to increase yields and thus raise margins. Intel plans to ship the next-generation of its Pentium chip, codenamed 'Prescott', at 90nm this year, and has publicly said it will do the same with its Pentium M mobile chip. AMD clearly needs to catch up.
In-Stat/MDR's mention of Paris is a reference to lower-cost version of the Athlon 64 with a quarter of the on-die L2 cache of the first Athlon 64s. They will have 1MB of cache, Paris will have just 256KB. Much less cache means a rather smaller die, and so more chips per wafer - again, it's all about punching out more processors for the same wafer cost, boosting margins.
Paris, incidentally, is a 130nm/0.13 micron part - its successor on the AMD roadmap is the 90nm 'Victoria'. Paris is due to sample early Q4 this year and go into full production during the same three-month period. Victoria is slated to appear late Q2/early Q3 2004. AMD is also expected to ship the 90nm version of ClawHammer, codenamed 'San Diego', in Q3 2004.
Opteron's success depends on cost, too. "AMD needs to make its Opteron family a success in order to improve its overall ASP to the point where it can be profitable," says In-Stat/MDR's report, the clear implication being that Opteron is currently losing AMD's money. IBM's support for Opteron, in the form of the eServer 325, isn't enough - AMD needs wider support from IBM, and from other tier one vendors.
In-Stat/MDR estimates that AMD grew its market share in x86 processors from 16 per cent in 2000 to 20 per cent in 2001, but dropped back to 16 per cent in 2002. It's leaving more up-to-the-quarter data for folk willing to buy the report, but we'd point out that Mercury Research's latest numbers put AMD on around 15.7 per cent.
Mercury puts VIA and Transmeta's most recently quarterly share at around 1.8 per cent, up on the "more than one per cent" they shared in 2002, according to In-Stat/MDR.
The latter's message for VIA is to reduce its dependence on Intel system buses. "Just as AMD did with Athlon, VIA must create an infrastructure independent of Intel's," it suggests, noting that VIA is well-placed to do so thanks to its expertise and experience in chipset design.
As for Transmeta, the prognosis is not good. It's lost a lot of support to Intel's Pentium M, and desperately needs to win erstwhile Crusoe customers back. Its TM8000 sounds impressive, but Transmeta is going to have a tough time pitching its against the 90nm, 2MB cached next-generation Pentium M, codenamed 'Dothan'. Winning back business will centre on keeping prices down, but that's not a good strategy if you're out to improve your bottom line, as Transmeta needs to do. In-Stat/MDR doesn't say so explicitly, but its tone suggests it doesn't believe the company has much of a chance.
"VIA's market share should remain stable in 2003 and 2004," it says, "but Transmeta will lose share during the product transition from the TM5800 to the TM8000."
In-Stat/MDR's report, Intel x86 Competitors, is available now for $2495. ®