April 27, 2006 1:04 PM PDT
Intel steps up chip cadence
- Related Stories
AMD server market share rolls onApril 25, 2006
Intel scrapes by in first quarterApril 19, 2006
Intel CEO throws down gauntlet to AMDMarch 7, 2006
Bruised Intel focuses forwardMarch 7, 2006
'Rev F' ends single-core Opteron era for AMDMarch 6, 2006
Intel's Core Duo health causes PC industry painJanuary 27, 2006
Weak demand, low prices ding Intel earningsJanuary 17, 2006
At a shareholder meeting Thursday, amid a discussion of restructuring and cost-cutting, the chipmaker disclosed plans to overhaul its underlying processor designs more than twice as fast as it has in the past.
Intel will adopt new "microarchitectures" every two years, Chief Executive Paul Otellini said at the company meeting. A microarchitecture is reused to deliver several ever-faster generations of a processor family. That stepped-up release rate will match the speed with which Intel moves to more-advanced silicon chip manufacturing processes.
"Every two years, we will bring out new silicon and new microarchitecture, instead of our four- to six-year cadence," Otellini said. Parallel design teams will leapfrog each other to introduce the new microarchitectures, with the Core microarchitecture starting to arrive in June, the Nehalem generation in 2008 and Gesher in 2010, he said.
The change, which Otellini said the company began internally in 2002, is part of Intel's response to suddenly fierce competitive pressure and tarnished financial performance. The Santa Clara, Calif.-based company has been losing market share to Advanced Micro Devices, which makes the Opteron server chip and Athlon desktop chip. Those processors provide technological advantages over and lower power consumption than Intel equivalents, even though Intel is a generation ahead in its manufacturing process.
AMD's rise and slowing PC sales have posed problems for Intel, which has curtailed revenue estimates and now is dealing with "several million" processors worth of excess inventory. "We see a tougher 2006 than we thought a few short months ago," Otellini said Tuesday.
But the company already has embarked on a plan to reduce expenses: When Otellini released first-quarter financial results last week, he said Intel is taking a "wholesale look at the company over the next couple of months."
Faced with such a plight, it's natural that the company is trying to direct attention to the future.
"If things suck now, what else are you going to do? You have to put your best foot forward," said Illuminata analyst Jonathan Eunice. But Intel is far from feeble, he added: "A lot of people would pay good money to have Intel's problems."
Intel has been grappling to move beyond the ill-fated NetBurst microarchitecture used in its current Pentium 4 and Xeon processors. It hoped the design would lead to very high clock speeds, but instead it led to cripplingly high electrical demands and waste heat. A high-end dual-core Xeon running full tilt throws off 165 watts of heat, compared with 95 watts for Opteron. It's no surprise, then, that Intel's new focus is on performance per watt.
The first elements of the faster design cadence will appear this June beginning with Woodcrest, a dual-core server processor for the mainstream servers with two processor sockets. In July will come Conroe, for desktop PCs, and in August, Merom for mobile machines, Otellini said.
"We made this change in 2002 inside the company. What you see now is the change with the manifestation with all new products," he said. "This is the effect of the right-hand turn."
The chipmaker continues to stumble
as AMD moves in on its market share.
Here's a look back at Intel's efforts to
overcome its losses:
Intel to get widespread
Mercury Research analyst Dean McCarron, who tracks processor market share, wouldn't predict whether the company will retake share from AMD. But the change should help Intel, he said.
"This is the first major architectural change they've done in about five years, and it should result in them being significantly more competitive," McCarron said.
The microarchitecture release schedule will be staggered with the schedule for new manufacturing processes, Otellini said. For example, the current process, which has circuitry elements measuring 65 nanometers, was introduced in late 2005; the Core microarchitecture will debut in 2006. The 45-nanometer process is slated for 2007, the Nehalem microarchitecture in 2008, the 32-nanometer process in 2009, and the Gesher microarchitecture in 2010.
When new manufacturing processes arrive, the existing microarchitecture will be adapted to produce a smaller version of that chip generation. For example, the Core microarchitecture, which began on the 65-nanometer process, becomes one code-named Penryn when it's geared for the 45-nanometer process.
But introducing new microarchitectures is tough. Some believe future generational shifts will be much less radical than the current shift from NetBurst to Core.
"Microarchitectures are very difficult things to design. There's a reason it takes years to do it," McCarron said. "I suspect what we'll end up seeing, with new microarchitectures every two years, is that the amount of change between the microarchitectures will be more modest, and there will be significantly more borrowing between the microarchitectures."
Intel, while not able to turn on a dime, has been relatively responsive to one AMD advantage, the move to dual-core processors. Intel's first dual-core models often put two last-generation chips into a single package, an inelegant but workable design.
"It was a little bit of oops, full stop, reverse course, but they did shift pretty quickly" to dual-core designs, Eunice said. "Some of the early steps look a little icky--this is glued onto that like with Pentium D--but they did get it out pretty quickly."CNET News.com's Candace Lombardi contributed to this report.