Intel bolsters Moore's Law with multi cores

Intel's solution is 'performance through parallelism'

Intel this year marked the 40th anniversary of Moore’s Law by elevating it from well-known maxim to the status of corporate mantra.

Moore’s Law has always been closely identified with Intel. Gordon Moore was CEO of the company when he coined his famous saying in 1965, noting that the number of transistors on a chip doubles about every 18 months.

The trend has remained roughly accurate since then, and despite occasional forecasts that Moore’s Law can’t be sustained, Intel executives are confident it will remain accurate for at least the next two decades.

“That law really, if you look at it, is the foundation of what we are all about,” current CEO Craig Barrett said in his keynote at the Intel Developer Forum this month. “If you want two words to describe our industry, it’s ‘Moore’s Law’, or ‘innovate and integrate’.”

Keeping pace with Moore’s Law meant that company had to abandon its fixation on higher clock speeds, with the attendant problems of heat and leakage. Intel now favours multiple cores and multiple threads to improve performance. In a later keynote, senior vice president Pat Gelsinger said Intel’s solution was “performance through parallelism”.

The industry has probably tracked a little behind Moore’s Law over the past few years, Gelsinger says, but he expects Intel’s multi-core processors to change that. The next four years, he says, will see “about the fastest rate of performance improvement of the past 20 years”.

Intel will be offering multiple-core processors this year for servers, desktops and laptops, according to Stephen L Smith, vice president of the company’s digital enterprise group. “Dual core will start at the top of our road map and become more mainstream over time,” he says.

By the end of next year 85% of the company’s server processors and 70% of the desktop and mobile chips will feature multiple cores, he says.

Two dual-core processors will be available for desktop systems this year, Intel says. The 90nm Smithfield chip will be marketed as the Pentium D processor, and a variant that also includes HyperThreading will be offered as the Pentium Processor Extreme Edition. Because HyperThreading offers two instruction threads per core, the Extreme Edition will have four threads per CPU.

A dual-core 90nm Itanium chip for servers, codenamed Montecito, is scheduled for release in the fourth quarter of 2005. It will also offer four threads per CPU. Two Xeon server processors, supporting x86 instructions with 64-bit extensions, will be released in dual-core 90nm and 65nm configurations early next year.

The first dual-core mobile processor, codenamed Yonah, is expected to ship this year, but Intel says the official introduction and “volume production” will begin next year. Yonah will be based on 65nm process technology. Smith expects Yonah’s power demands to be similar to today’s single-core Pentium M chips.

Smith says single-core processors will remain in Intel’s lineup while the multi-core chips are rolled out.

Intel says it has many dual-core and multiple-core designs in development, although it won’t predict when its multiple-core processors will reach the market.

And what of the years following 2025, the point where Intel’s own forecasts suggest physical limitations will prevent Moore’s Law-rate performance increases? Barrett says Intel can imagine getting transistor sizes down to about 5nm — about ten times the space between individual atoms.

“Beyond that, leakage currents get in the way and a bunch of other things,” he says.

“But there’s lots of life left in this technology and every time we seem to run into a roadblock, the bright technologists and engineers that we have, the R&D funds that we make available to them — they always seem to circumvent those roadblocks.”

Cooney attended IDF ’05 as a guest of Intel

Tags intel

More about Intel

Comments

Comments are now closed

Wynyard to support NZ Police's sex offender register

READ THIS ARTICLE
DO NOT SHOW THIS BOX AGAIN [ x ]