Intel on Friday chose to scrap part of its existing processor roadmap, dropping single-core processors in favour of chips with two cores.
The decision will help the company improve the performance of its future chips without having to rely on a power-centric design, analysts said.
Two future processors were eliminated from Intel's product plans: Tejas, once the future of the Pentium 4, and Jayhawk, a planned successor to the Xeon server processor. Instead, Intel plans to roll out dual-core designs across all of its server, desktop and mobile processor lines by the end of 2005.
Dual-core chips offer two processor cores on a single die. Chip designers have been gravitating toward this design because it allows them to use two lower-power cores to improve performance rather than depending on a high-frequency single core.
Nathan Brookwood, principal analyst with Insight 64 in Saratoga, California, drew an automotive analogy for the shift to dual-core processors.
"This is a shift from a one-cylinder engine to a two-cylinder engine. It runs smoother, it gets better mileage, and you can use two smaller cylinders instead of one huge cylinder," Brookwood said.
IBM and Sun Microsystems have each released dual-core chips for high-end servers. Intel has also announced plans to bring dual-core chips to its Itanium server processor family.
The dual-core design lets Intel handle a power-consumption problem that was coming to a head with the February release of the Prescott Pentium 4, the company's first chip built on a 90-nanometer process generation. Power consumption is of paramount concern to chip designers as circuits shrink to the point where power can leak out of those circuits as heat.
In almost every other transition between process generations, Intel has been able to follow a simple plan for managing frequency and power consumption, said Kevin Krewell, editor in chief of The Microprocessor Report.
The company would steadily increase the frequency of its older chips to the limit imposed by the older process generation. Once the new process generation was ready, Intel would introduce new chips that run at even higher frequencies but with lower power consumption due to the benefits of the new process technology, he said.
However, Prescott actually consumed more power at the same frequency of its Northwood predecessor. This was due in part to new instructions and additional cache, but analysts also feel that power leakage is worse than expected at the 90-nanometer process generation.
Each core on a dual-core processor can run at a lower frequency, and therefore consume less power, while still delivering better performance than a higher frequency processor with a single core, Krewell said.
Intel's President and Chief Operating Officer Paul Otellini told analysts last year that the Pentium 4 would hit 4GHz by the end of 2004, but Prescott's power consumption indicates that scaling beyond that frequency would be extremely difficult, Krewell said.
While the power benefits of dual-core chips are clear, their performance depends on the availability of software that can take advantage of the independent cores, said Gordon Haff, principal analyst with Illuminata.
Desktop applications have traditionally been single-threaded, Haff said. A software thread is a stream of instructions that can run independently of other instruction streams.
The advantage of a dual-core processor is that it can run a multithreaded application much faster than a single core processor because independent software threads can be processed at the same time, Haff said. But a single-core processor running a single-threaded application will almost always be faster than a dual-core processor running a single-threaded application, he said.
Intel has already helped the development of multithreaded applications with Hyper-Threading, its technology for allowing desktop and server processors to take advantage of unused execution units in those processors by making the operating system think the computer has two separate processors.