Intel show to tackle debate over dual-core chips

At an Intel-sponsored developer conference next week, one of the big debates will be, "what exactly is a dual-core processor?"

The Santa Clara, Calif.-based chipmaker will provide details about a host of processors at the three-day Intel Developer Forum in San Francisco. On tap: a discussion of chips for servers, notebooks and desktops coming out later this year that will contain two computing cores instead of just one. Adding cores lets computers handle multiple applications simultaneously and complete time-consuming tasks more rapidly.

The definition of "dual-core processor," however, is broad and ranges from a single chip in which the two integrated cores share resources to something that consists of two functionally and physically separate pieces of silicon that happen to be in the same package. Packaging, which protects the silicon and contains the metal bumps for carrying signals from the chip to the rest of the computer, is an integral part of a processor.

Intel processors will span the spectrum. In Montecito, a dual-core Itanium chip, the two cores under the same roof will each have a dedicated memory cache for rapid data access. The two cores, however, will share a bus for shuttling data to the outside world, as well as other resources, such as an integrated component, code-named Foxton, for saving energy.

By contrast, the Smithfield chip for desktops is more like a condominium. The two cores will come on the same piece of silicon but will largely function independently, according to sources familiar with the company's plans. Until somewhat recently, Intel debated making Smithfield out of two separate pieces of silicon in one package, sources said. It is expected to run around 3GHz to 3.2GHz.

This expected lack of intimacy between the cores in Smithfield is probably beneficial to Intel, according to analysts. It is likely one of the reasons the company was able to advance the commercial release of dual-core desktop chips from 2006 to the second quarter of 2005, several months ahead of the first desktop dual core from rival Advanced Micro Devices.

Consumers won't experience a difference in performance between a highly integrated dual-core processor and two chips that simply share the same package. Nonetheless, a comparative lack of integration will reinvigorate the debate over whether Intel or AMD is further along in dual-core development. In the past few years, AMD has touted ideas such as 64-bit computing and faster input-output links first, putting Intel

in the uncomfortable position of a follower.

Like Montecito, AMD's chips will have dedicated caches, but the two cores will share things like a memory controller.

"I would give AMD an edge in integration," said Kevin Krewell, editor in chief of the Microprocessor Report. Still, by the time AMD's desktop chip hits, Intel will have an integrated dual-core notebook chip and a new dual-core desktop chip called Presler.

Coming out with a dual-core Smithfield that consists of two separate pieces of silicon would be a publicity headache, Krewell added.

"Even though functionally it would be identical, in terms of silicon manliness it would be cheating," he said. "It would also sort of weaken their case with Oracle on licensing."

Intel and other hardware makers are worried that traditional by-the-processor models for software licenses will restrict sales of their multicore offerings. Oracle, for instance, considers dual-core chips as two processors. That means customers wind up paying more for software that runs on dual-core systems. In contrast, Microsoft said its licensing terms will count dual-core chips as a single processor.

An Intel representative declined to specifically discuss Smithfield but said the company plans to use an array of design techniques to bring dual-core chips to different markets.

Unclogging data traffic
Another highlight of the conference will be the first discussion of the Intel I/O--for "input/output"--Acceleration Technology designed to unclog data traffic that's transmitted with the Transmission Control Protocol/Internet Protocol. Ideally, the technology will speed up the interactions between PCs and servers by up to 30 percent. The software and specially tweaked networking chips to enable the technology will arrive in 2006.

The conference officially begins March 1 with a keynote speech from Intel chief Craig Barrett, who will step down as CEO to become chairman in May. Incoming CEO Paul Otellini is not speaking but will attend the conference.

Other keynote speeches will come from the general managers of the new platforms groups. Rather than make individual products, such as chipsets or flash memories, the platform groups seek to develop an array of chips, boards and reference designs to sell to business customers, the communications industry or consumers. The idea is that selling bundles of integrated chips can bring more revenue than selling single chips. Texas Instruments, Samsung and others are taking a similar approach.

Speakers include Pat Gelsinger, co-general manager of the Digital Enterprise Group; Don MacDonald from the Digital Home Group; Sean Maloney, co-general manager of the Mobility Group and an oft-mentioned candidate for the Hewlett-Packard CEO slot; and Justin Ratter, an Intel senior research fellow.