Dean's World

Moore's Law, often stated loosely (and somewhat inaccurately) is that computers double in power and halve in price about every 18 months. This is at best a "rule of thumb" definition that describes something much more complicated: that the price/performance ratio of technology has been on an exponential growth curve for some time now, and shows no sign of stopping.

If you apply the rule haphazardly, you would say that, for example, in 20 years or so you should be able to find computers running 30,257 gigahertz as opposed to the 3 gigahertz machines you can buy today. But as I noted in the comments to an earlier article on this topic, that exercise is more about imagination than reality, because "hertz" isn't a realistic estimate of computer speeds, and it's not the only part of performance anyway; it's just that running that sort of calculation gives you a "horseback guesstimate" of just how fast the growth curve in technology is.

In truth Moore's Law doesn't stick to the 18-month figure. Indeed, Moore himself originally gave a figure of 24 months, and that was purely about transister density on chips. But the "18 month" figure has come into general use mostly to describe a very loose rule of thumb about computer technology improvements in general.

Computer clock speeds are measured in "hertz," i.e. cycles per second. In 1985 I had a computer that used an 8-bit, 1 Megahertz microprocessor called the 6502. This is what the growth curve on clock speed has actually looked like over the last few decades, and where it's projected to go over the next 10 years or so:

(I snitched that from Ray Kurzweil. See below.)

Yet today if I could scale that old 1 Mhz computer up to where it ran at 3 Gigahertz (anyone else out there want to imagine a 3 gigahertz Apple //e?), it would seem very slow to modern standards. The tasks the modern Pentium and Athlon chips can do running at the same 3 Gigahertz are much, much greater, due to many tricks that chip designers have created over the years: the ability to move larger amounts of data per cycle, to run multiple instructions in a single cycle, parallel processing, specialized types of processing using more advanced math, even anticipated processing, where the processor can usually correctly guess what calculations are needed before they're asked for. All of these are different measures of performance that can't be calculated in a simple "hertz" rating.

Most technological advances in computers have exponential growth curves over time. But disk storage increases at its own rate. Network bandwidth increases exponentially, but at a different rate. Prices decrease exponentially, but at their own rate. And so on. Due to short-term disruptions sometimes you see temporary hitches in the growth rate of all of these, but they've always so far broken past their limitations either by new breakthrough improvements (different materials for CPU chips, for example) or obsolescence (I doubt if my son will remember floppy disks).

Yet all these technologies--and a good many non-computer technologies, including biotechnology and nanotechnology--have been on exponential growth trends for some time now, and none of them show any sign of seriously stalling out.

I suggest that if you doubt it, I suggest you read Exponential Growth an Illusion?: Response to Ilkka Tuomi by Ray Kurzweil.

(By the way, the more I learn of Kurzweil's work, the less goofy he seems.)