Fabless chip companies are flawed, AMD claims
Rise or fall?
The going will be tough for new fabless CPU companies –- Rise, Transmeta and others of that ilk –- predicts Dana Krelle, AMD vice president. Fabless companies are operating a flawed business model, which will turn them into bottom feeders, he claims. “Only by innovating ahead of Intel are you able to add value,” he says. Fabless chip companies will be unable to do this, he says. “The trouble with foundries is that they want their 50 per cent profit margin,” he says. Throughput at contract foundries is rarely at optimal capacity, he says. And operators do not supply access to bleeding edge technology at the design stage. The upshot is that fabless chip design companies into the unprofitable, down market sector. “It’s a business that’s damn hard to survive,” he says. Krelle bases his judgement on his previous life with Nexgen, a CPU house more or less forced into merger with AMD a few years back. “There used to be two big fabless chip companies – Nexgen and Cyrix – where are they now? Nexgen is part of AMD and Cyrix is part of National Semiconductor.” Speaking at Comdex, Krelle argues that AMD has pulled far away from the rest of the pack – including NatSemi. “Intel and AMD are the only two companies that share both following characteristics: ownership of foundries and multiple, top-notch design teams that keep innovating on a multi-generational basis.” The inference we are to draw is that NatSemi/Cyrix is hamstrung because it works on single lifecycle upgrades. With the introduction next year of the K6 Sharptooth and the K7, is ready to take on Intel in all market segments, Krelle claims. Operating at 400MHz clock speed, Sharptooth outperformed the Intel Pentium II 450MHz in ZD Winbench tests, in a demo shoot-out arranged for The Register’s benefit at AMD’s meeting rooms in the Las Vegas Convention Centre. “We’re not optimised for clock, we’re optimised for performance,” Krelle said, in what sounds like a well rehearsed soundbite. At one point Atiq Raza, AMD’s co-chief operating office and chief technical officer, ambled in, explained the company’s difficulty in positioning Sharptooth against Katmai, and then ambled out again while discussing an email concerning a meeting Microsoft with an underling. Now what could that be about, we wonder? Sharptooth is pitched at the performance segment of the market – and will take into the high-end consumer business, Krelle says. He rejects suggestions that Sharptooth is a short shelf life transitional technology, bridging K6-2 and K7. When the chip moves to .18 micron level production at the end of 1999 t becomes feasible that Sharptooth becomes the entry-level product,” he says. K7 will fit inside PCs in the $1,995 -$2,995 bracket, Krelle revealed. The K7’s strong floating point performance will take it into the graphic intensive market. We imagine that Intel-hater Intergraph will be among the first in the queue to place their K7 orders. In a demo, Krelle put the K7- powered PC through its paces, in which a DVD software version of Godzilla the movie was played. Graphics performance was certainly impressive. And launch date for the K7? “This is first silicon,” Krelle says. “We’re on track (for shipping) for the first half, next year.” It’s a tough life being the Avis of the computer industry. Trying harder in AMD’s case means having to second guess Intel on technology and market positioning fronts. Pricing is not such a big issue, according to Krelle, secure in the knowledge that any price war would hurt Intel more. Wall Street punishes any Intel failure to meet the numbers very severely. “It is a fine outcome when a market ends up with two companies, both of which are nicely profitable,” Dana Krelle, AMD says. “That’s the way it works in many industries.”
It all sounds very cosy. We’ll just have to wait for next year’s publication of AMD’s deposition in the Intel FTC anti-trust case -- to bring out the attack dog in the company. In the meantime, AMD is contenting itself with playing hardball on the technology front. ®
Sponsored: Benefits from the lessons learned in HPC