Goodbye Gigahertz

by on October 15, 2004

The GHz race officially came to an end this week. No, really. Intel, who has held the speed crown for more than 5 years, has thrown in the towel, announcing that they would break the 4 GHz barrier… well, never.

This is a development that analysts have been predicting for years. Since the late ’90s, CPUs have been much faster than the memory and buses that feed the CPU with data. That means that more processor speed is mostly useless for the vast majority of data-intensive tasks. Worse, Intel cheated in designing the Pentium 4, ramping up the speed mostly by reducing how much the chip did on each cycle. The result was a chip that had a higher clock rate, but didn’t actually perform any better than slower-clocked chips that did more with each cycle.

The design of the Pentium 4 was driven by marketing, not engineering, considerations: GHz was an easily understood metric for judging processor speed, and so having the fastest chip was an effective selling point. But the reality has become so obvious that even marketing people can’t ignore it. If they had continued on their current path, they would have needed ever-more-elaborate cooling technologies to keep the chip from melting.

From now on, expect chipmakers to focus on greater parallelism– putting more than one processor on a chip, executing multiple instructions per cycle– and on non-performance features like reducing power consumption. Both IBM’s G5 (which is in Apple’s new Power Macs) and AMD’s x86-64 architecture do a better job of getting more performance out of fewer clock cycles. The shift to non-performance-related features is already apparent with Intel’s Centrino line, which is targeted at mobile devices and boasts lower power consumption and wireless features.

Comments on this entry are closed.

Previous post:

Next post: