Barrett Lays Out Intel's Battle Plan

Electronic News, Sept 21, 1998 by Peter Brown

Although Intel this year has seen competitors make deep inroads into the low-end desktop market, and has been criticized as being ill-equipped to compete in markets outside the PC space, Dr. Barrett made it clear he was focused on the big picture. ``We see there being one billion computers in the world by 2005,'' he told the Intel Developer Forum (IDF), and made it clear Intel was prepared to fight to be in every one of them.

Key to keeping its 85% MPU market share over the next couple years is the Katmai processor, aimed squarely at the mainstream PC space. Scheduled to begin production in the first half of 1999, Katmai will boast a performance of 450 to 500MHz, supported by the 2x AGP 440BX chipset, he claimed.

Katmai's performance will be helped by 70 new instructions per processor, AC-3 audio, and MPEG 2-enabled encode/decode integrated on the processor. Other specs include a 2 gigaFLOPS per second peak performance using a 128-bit SIMD architecture and 512 kilobits of BSRAM on-chip. It will be manufactured on a 0.25-micron process.

Dr. Barrett said Intel will follow that in the second half of next year with the Coppermine processors for both desktop and mobile computers. He gave few details, other than saying Coppermine would be built on a 0.18-micron process.

The Mobile Pentium II processor will get a speed boost to 333MHz in the first half of 1999 utilizing the 440BX chipset. In the sub-$1,000 space, where Intel has come under intense pressure from Advanced Micro Devices and National Semiconductor, the company is planning to introduce a 366MHz Celeron processor in the first half of 1999, with unspecified upgrades later next year.

Intel will unveil StrongARM processors in the first half of next year for the set-top box and handheld PC market. The company has been a fringe player in this market, and is banking on the StrongARM 1100/1500 processor to jump-start its fortunes. Follow-on processors are slated for the second half of the year.

Over on the server/workstation side, Intel is planning on rolling out the follow-up to the Xeon, dubbed Tanner, in 1H99. Built on a 0.25-micron process, Tanner will feature a performance of 500MHz, L2 cache of 512K, while continuing to use the 450NX and 440GX chipsets. In 2H99, Intel will debut the Cascades processors for this space, based on the Katmai architecture and using a 0.18-micron process.

As far as its much-anticipated Merced chip goes, Dr. Barrett said the 64-bit line of processors are right on-track for an initial rollout in mid-2000. The company said the logic design has been completed, the circuit design and layout are currently being taped out and OEM platform designs are on-track for first samples in the near future.

Merced is aimed at enterprise servers and workstations.

In addition, the company has plans past the Merced for a chip called McKinley that is targeted for production sometime in 2H01. The McKinley chip is planned to double the performance of the Merced chip. Intel continues to promote the IA-64 architecture as having explicit parallelism, massive resources, predictabiity and speculation capability.

Demise overstated

Despite its troubles on the low end, Intel remains formidable, said Scott Hudson, analyst at Cahners In-Stat Group in San Jose, Calif. "With tremendous fab capacity, 85% market share (in microprocessors), and probably one of the biggest marketing muscles in the world, the company is going to survive. This is not to say there won't be tough times for the company, but everyone has tough times. They are and still will be the leader in PCs."

No Immediate Copper Plans

Although some major players like IBM and Motorola have detailed their copper metalization plans, Intel has kept a tight lid on its intentions. Dr. Barrett, however, said the company's move to copper may not be as quick as some of the others. Instead, it will concentrate on what it considers poor transistor performance.

The real bottleneck, he says, lies at the transistor level. "We will be as fast with our aluminum metalization and fast transistors as others will be with copper technology," Dr. Barrett said. Decreasing the gate delay that increases with each process geometry shrink is key, he noted. But he didn't rule out using copper in future process generations.