Back then, Intel lorded over AMD. Craig Barrett, who succeeded Andy Grove as Intel CEO in March 1998, kicked off a strategy to diversify the company's revenue base.
AMD, meanwhile, was struggling with financial losses and sporadic inventory and manufacturing problems. Analysts questioned whether the company would have to find co-tenants to help pay for its planned fabrication facility in Dresden, Germany.
In the background, seemingly innocuous events began to transpire. In the middle of 1998, Intel halted construction on a plant in Fort Worth, Texas, because of the tepid economic outlook in Asia and the failure of a tax reform bill in the Texas legislature, according to Intel spokesman Chuck Mulloy.
Around the same time, Intel engineers in Folsom, Calif., were discovering that the upcoming 820 chipset for faster Pentium IIIs (code-named Coppermine) and a then relatively obscure memory technology called Rambus weren't working together as planned.
"Rambus clearly was more difficult than people expected," recalled Nathan Brookwood, an analyst at Insight 64. "Intel had to go back and learn about how to channel these high-speed signals on a motherboard...They were getting all sorts of subtle problems. They had to substantially upgrade the equipment they had in their labs and rethink how they did testing."
Meanwhile, one exit north on Highway 101 in Sunnyvale, Calif., AMD was working on a chip called the K7. Dirk Meyer, a Digital Equipment alumnus and director of engineering at AMD, managed the project, which he would describe at the Microprocessor Forum in October.
How times change
Flash forward to the first half of 2000. Intel found itself mired in a dire processor shortage caused in part by a series of delays in 1999 of the 820 chipset and other parts. The scrapping of the Fort Worth plans also was having an effect. "That decision has had some consequences this year," Mulloy said.
In addition, Rambus memory had failed, as predicted, to come down in price. To spur sales of Coppermine Pentium IIIs, Intel came out with a bridge chip called the Memory Translation Hub (MTH) that was supposed to allow Pentium III PCs to use cheaper standard memory. An MTH-related recall cost Intel $253 million and, more importantly, gave a black eye to its sterling reputation for manufacturing.
In June, Intel canceled Timna, an inexpensive processor that would have been paired with the MTH.
Non-Rambus problems cropped up too. The release schedule for Itanium, a 64-bit server chip destined to take on Sun's UltraSparc processors, was pushed back to 2001. The 1.13-MHz Pentium III was recalled.
"In a little bit of our haste or excitement to get out the product, we didn't go through all of our traditional product validation," said Jeff McCrea, marketing director for Intel's desktop products group.
The company managed to come out with the Pentium 4, perhaps the bright spot of the year. Still, partly because software has yet to be tweaked for the chip, benchmark tests have shown the chip to provide middling performance improvements.
By contrast, AMD was seeing tremendous consumer acceptance with Athlon, the public name for the K7. With the new chip, AMD took the speed crown from Intel, landed contracts with all the major PC makers but Dell Computer, and obtained a new reputation for high-performance technology.
It also didn't face shortages to the same degree. For the first time since 1995, the company will post an annual profit in 2000.
While AMD's market share hasn't radically improved, its place in the market has moved. Rather than being confined to the low end, the company now has a spot in the more lucrative performance segment. In the U.S. retail market for PCs in the $1,000 to $1,500 range, AMD saw its market share go from 3.3 percent in May to 22.8 percent by September, according to figures from PC Data.
Key to the company's success was improved manufacturing. Historically, AMD has flubbed chip launches, leaving them vulnerable to Intel price cuts. "They went from having a product they couldn't build to having a product they could," said Dean McCarron, principal analyst at Mercury Research.
The credit for turning around the manufacturing situation belongs largely to Bill Siegel and Jim Duran, two AMD executives, said Kevin Krewell, an analyst at MicroDesign Resources and a former AMD employee.
Intel has always prided itself on a manufacturing methodology called "copy exact" in which each fab is identical to the next, allowing the company to accelerate manufacturing on a new chip. "AMD had a process called 'somewhat similar,'" Krewell joked. Siegel, a former IBMer, began to correct AMD's situation in the mid-1990s, but the real benefits of his work came to the fore with Athlon.
Duran, meanwhile, worked at both the Austin and Dresden facilities to see that the manufacturing improvements were carried out. "He was sort of the behind the scenes guy who ironed out the problems at Austin," Krewell said. "It was really those three guys," Meyer, Siegel and Duran.
Additionally, some of the credit for the turnaround, Brookwood said, goes to Atiq Raza, AMD's former president, who left in 1999.
Brookwood also gives credit to the new German plant. "It was a piece of luck that (Athlon) came out just when the industry had a capacity shortage and they had a product the industry wanted," Brookwood said.
In addition, the Dresden plant allowed AMD to shift relatively painlessly to producing chips with copper, rather than aluminum, circuits, a migration that has caused problems for some other manufacturers. Intel will make the copper jump this year.
"What they are praying for is that PC demand is less than expected," said Nimal Vallipuram, an analyst with Dresdner Kleinwort Benson.
The two will also face new challenges in notebooks. Transmeta burst onto the scene with its Crusoe chips, which consume less power than other chips, according to the company. Low power consumption became a reigning theme in notebooks.
Who will win? Stay tuned for the next two years.