Intel, which next week is expected to announce plans to move to a new processor architecture, is switching to a new yardstick to measure processor performance: performance per watt.
Intel Corp. is expected to detail next week at its IDF (Intel Developer Forum) a plan to begin building multicore chips with the architecture, a modified version of the circuitry behind its Pentium M notebook processor, during 2006.
Intels announcement will publicly signal an internal shift thats already taken place. After years of promoting clock speed, its now emphasizing overall performance and power-efficiency.
Intels shift to processor numbers and its wholesale move to multicore processors—a multicore chip includes pack two or more processor cores in one package—sealed the deal for the architectural change, as power can be a limiting factor in fitting two or more processor cores together into a single chip.
Chips with multiple processor cores boost PC performance versus single-core chips by splitting up jobs.
“Its not even so much even performance per watt as it is fitting higher performance computing into more constrained environments, either constrained by power, by [heat] or by noise or size,” said Dean McCarron, principal analyst at Mercury Research Inc.
Particularly for desktop PCs, “Its all about attaining the maximum performance you can in an environment that, unlike in desktops of the past, now has some constraints on it. As you move to multiple-core devices, scaling the frequency higher isnt as important as the ability to put multiple cores on a chip, anyway.”
Thus the performance-per-watt plan was born and Intels Israel-based processor design team, which created the Pentium M, using Intels P6 architecture a base, appears to have gotten the job.
As part of the shift, Intel also cancelled Tejas, a speedier successor to the Pentium 4.
Although Tejas would have been faster than the current Pentium 4, potentially offering 5GHz speeds, it would also have been more power-hungry.
In contrast, dual-core or multicore chips based on the new architecture can, in theory, be made to offer competitive performance with lower power consumption.
Although it does not run as fast as the Pentium 4 chip and lacks features such as 64-bit addressing, a given Pentium M uses about a quarter of the power of a similarly ranked Pentium 4 and performs nearly as well.
Thus Intel is expected to create the equivalent of a super Pentium M with some features that are also found in NetBurst, including Hyperthreading and 64-bit addressing.
The resulting chips are expected to be faster and to have more capabilities than the Pentium M, but be just as miserly on power.
These power-efficient, multicore chips will bump up the performance of notebooks and servers. But theyre likely to have to most visible and drastic affects on desktops.
Analysts predict they will help spell the end of the desktop towers for most applications, spawning instead smaller, quieter machines that are not unlike Apple Computer Inc.s Mac Mini.
Ultimately, these miniature desktops will evolve into high-performance hubs for connectivity, attaching to the backs of flat panel displays or hiding out of the way in a living room cabinet, while still providing basic functions such as network access and storage, analysts predict.
Intel has already taken some steps in this direction by promoting desktops that use its Pentium M processor.
Its first dual-core Pentium M, a chip dubbed Yonah, will become part of Intels forthcoming Napa notebook platform, due in early 2006. But it will also be used in brand-name desktops, possibly including the Apples Mac Mini, said sources familiar with Intels plans.
Merom, a dual-core notebook processor that will succeed Yonah in late 2006, will be one of the first new architecture chips.
Expected to arrive in notebooks in early 2007, it will deliver the expected bump in speed and 64-bit addressing, the sources said.
Intel, which had kept the Pentium M separate from NetBurst-based Pentium 4s and Xeons, will build future desktop and server chips on the new architecture as well.
Thus Intels Conroe chip, expected to be the first new architecture desktop chip, will share the same circuitry as Merom and Woodcrest, a forthcoming server chip.
The chips will be tuned for their respective applications, however. Woodcrest, for one, will include specific enhancements for servers.
The availability of a high-performance, but power-efficient processor such as Conroe—or possibly Merom, which might show up in some desktops just as the Pentium M will—could almost completely change the desktop PC, said Roger Kay, president of Endpoint Technologies Associates Inc.
Kay envisions the desktop becoming “a connector for high performance, high comfort peripherals,” he said.
“Notebooks will use the same technology, but the monitor is integrated. Comfort and performance is the desktop story, and mobility is the notebook story. But, other than that, theres no reason to have them be that different.”
Given Intels emphasis on dual-core processors—it has more than a dozen such chips in the works—and a similar maneuver by its rival Advanced Micro Devices Inc., choosing a more power efficient path made the most sense, McCarron said.
NetBurst development will cease, but the architecture is expected to live for several more years.
Pressler, a 65-nanometer version of the Pentium D thats due early next year, is likely the end of the line for NetBurst. The new architecture chip Conroe will ultimately replace Pressler.
“Intel is picking one of the two architectures it has and going with it,” McCarron said. He predicted that NetBurst chips wouldnt disappear until around 2008.
Aside from discussions of its dual-core and multicore processor plans—2006 will be the year in which most Intel chips move to dual cores—as well as its next generation architecture, Intel will also detail its latest platforms plans at the IDF, next week.