copyright 2004-2008 by Nicholas G. Carr
site design by timothy swan
Nick Carr's essay blog
October 17, 2004
Intel’s retreat continues. In November 2003, the giant chipmaker announced in a webcast that it would introduce Pentium processors with clock speeds of four gigahertz by the end of 2004. In July, it pushed the target date back to the first quarter of 2005. Last week, it quietly confessed not only that it would miss the revised deadline but that it was discontinuing plans for the 4 GHz chip altogether. Rather than seek further boosts in CPU clock speed, Intel said it would improve processor performance by focusing on secondary factors: increasing bus speeds, enlarging memory caches, and ultimately producing dual-core microprocessors - chips with two processing engines.
Intel is not alone in struggling to achieve further clock-speed gains. AMD, Motorola, and IBM, among others, are also having problems. In fact, the clock-speed curve, which shot breathtakingly upward during the nineties, has been flattening for some time now. The immediate cause is technical. As chip speeds accelerate, it becomes ever more difficult to keep power consumption and operating temperature under control; reliability and usability both begin to suffer. At the same time, the manufacturing margin of error narrows, and it becomes harder to keep the yield of good chips up.
But Intel and other chipmakers are nothing if not resourceful. They’ve faced and surmounted tough engineering problems before. What’s really holding up progress today is not so much technological issues as economic ones. The demand for higher clock speeds is fading. Indeed, the processing power of mass-market computers long ago advanced beyond the need and the ability of most users to put that power to use. Sure, gamers and hurricane-modelers would love to have those 4 GHz chips. But at this point most users - particularly business users - really couldn’t care less. They have more than enough horsepower to do everything they need to do.
As demand for increased CPU power weakens, chipmakers’ incentive to invest in achieving higher clock speeds fades as well. It becomes ever more unlikely that they’ll earn enough new sales to reap any return on the investment required to achieve the next bump in speed. The economics just don’t work – and that’s the real reason Intel backed out of its 4 GHz promise. As one financial analyst put it in commenting on the company’s announcement, “the cost [of crossing the 4 GHz barrier] would far outweigh any financial benefit.”
If we take clock speed as a proxy for Moore’s Law – that microprocessor power goes up at an exponential rate, doubling every year or two – then the implication is clear: That most famous of all Computer Age prognostications is reaching its expiration date. The denizens of Silicon Valley will howl at such a statement. They’ll point out that, first of all, clock speed is not the same as processing power and, second, processing power is not really what Gordon Moore was talking about anyway. When he wrote the brief 1965 paper from which his law was later derived, he was making a series of arcane (if prescient) observations about chip design, manufacturing, and, most of all, cost. Clock speed hardly entered into it.
But that’s the point: Moore’s Law has always been a fabrication – a popular (mis)conception of reality rather than a precise description of reality. And it’s a fabrication carefully nurtured by Intel and other companies in the computer business. They’re the ones that promoted clock speed as a measure of “power” and, in turn, a general proxy for Moore’s Law. It was a very useful and lucrative fiction for marketing purposes. Your average joe’s eyes quickly glaze over at discussions of transistor density and feature size, but the prospect of going from 1 GHz to 2 GHz – well, that gets the old saliva flowing.
Or at least it did. As it’s become harder for average mortals to see the benefits of faster clock speeds and greater CPU muscle, the guns of the great megahertz war have begun to go silent. In the computer business today, the market places greater value on factors other than processor power – on things like battery life and portability and not having your laptop burn a hole through your trousers. As computing devices keep getting smaller, such secondary characteristics will continue to grow in relative importance.
Here’s the so-what: Whether you’re talking about planes, trains, or microchips, the basic performance of all technologies eventually reaches the limits of market demand - and then progress shifts gear. Innovation begins to focus less on what might be called core functionality and more on a myriad of secondary traits. Boosting the capacity or security of a railcar, say, comes to have more market value than increasing locomotive speed. When that point arrives, the success of a vendor hinges more on its ability to foresee which factors customers will value than on its raw technical skill.
In the end, it’s the customer, not the engineer, who determines the course of and the limits to technological progress. Moore’s Law may be expiring, in other words, less because of technical limitations than a simple lack of interest.