Intel Scraps Tejas, Will Focus On Dual-Core Processors
Intel has confirmed that it is scrapping development of Tejas, its next-generation chip, and will accelerate development of dual-core processors for use across its server, desktop, and mobile platforms next year.
Intel told its employees and customers that it will reassign resources from Tejas, the follow-up to its current Prescott version of its mainline Pentium 4 processor line, and concentrate instead on moving up the introduction of dual-core designs about 12 to 18 months from its original plans.
The decision appears to stem from continued difficulties associated with production of high-frequency processors in Intel’s new 90-nanometer manufacturing process and associated power-consumption issues. Processors with multiple cores would be able to operate at lower frequency while delivering increased performance.
“It seems like this is the way the marketplace is moving, and rather than being late to that, we believe we should do everything thing we can to accelerate development of dual-core and readjust the road map,” an Intel spokesman says.
The decision is identical to one made by Sun Microsystems a month ago to scrap development of its UltraSparc 5 chip and instead concentrate on the introduction of a four-core, eight-way threaded UltraSparc 4 design scheduled for introduction in 2006 that would effectively provide 32-way processing in a single chip, says Nathan Brookwood, an analyst with Insight64.
“Clearly this is an acceleration of Intel’s multicore strategy,” Brookwood says. “The question is whether they are accelerating the strategy because they thought its [multi-core] technology is looking better or because the other path was looking worse.”
Intel had already revealed plans to bring a dual-core Itanium processor to market next year and will add a dual-core product for the desktop and mobile markets, the spokesman says. The timing on dual-core Xeon chips remains uncertain, although that line, too, will move to dual-core by 2006, he says.
The decision is also in line with comments made by Pat Gelsinger, Intel’s senior VP and chief technology officer, at the Intel Developer Forum in February. He said the company will increasingly look to multicore processing as a way to improve performance rather than focus on increasing the clock speed in single-core products.
The company also recently disclosed its plans to stop marketing its processors based on clock speeds and instead move to a more-generic product numbering that would attempt to gauge overall performance on a combination of factors such as clock speed, bus speed, and on-chip memory.
Multicore processing systems have long been on the market, some with 100 or more processing elements. But they have had limited success. They generally use slower individual cores in combination to deliver increased overall performance.
The difference now, Brookwood says, is that multicore products such as the new Intel and Sun designs will address markets that already have applications that can readily take advantage of the multicore designs.
“This does look like a more-promising approach,” Brookwood says. “With more cores you can have the same number of transistors [as high-performance single-core implementations], but you don’t have to push them as hard on frequency and can use lower voltages.”
