CPU Performance

How far have we come?

  I have a little program, called the Autodesk Benchmark, that I run on various computers to get a feel for how fast they'll run engineering software like AutoCAD.  

Here's an anthology of results, spanning the history of Autodesk, from 1982 through the present.

In 1982, personal computers were becoming seen as useful tools for serious work, but they were still very, very slow for computationally intensive tasks. A typical low cost PC such as a Commodore took close to an hour to run the program. (Putting this in perspective, to do the same job by hand with a pocket calculator would probably take a whole day.)

More expensive and powerful PCs emerged and slowly reduced this time, making more and more complicated tasks practical. Then, in 1984, then 80286 appeared. The impact of this machine on Autodesk can't be underestimated. Calculations that took half an hour on the PCs that existed when we started the company could be done in a minute on the PC/AT. It's no coincidence that Autodesk's sales took off through the roof right about that time.

But it didn't stop there. Three years later, workstations and PCs based on the next generation of chips had cut this time by another factor of ten--from a minute to about five seconds.

And now the newest, shiniest crop of machines just arriving have handed us another factor of twenty--down to less than a third of a second.

In eight years, we've seen a task that originally took an hour reduced, by the simple consequences of device scaling--making things smaller--to less than a third of a second.

This kind of technological progress is hard to comprehend, even if you've lived through it. If automotive technology had advanced an equivalent degree, your car that went 55 MPH in 1982 would today go 615,000 miles an hour, with the same gas mileage and   price. You could drive to the Moon in around 25 minutes, if you beat the rush hour and had a good radar detector.

So the inevitable question is...just how long can this go on?

There are reasons to believe the end of progress in electronics through pure application of device scaling may be within sight, although not imminent. When the limits of device scaling are encountered, the linear extrapolation that has driven our industry since the 1950s will come to an end and we'll enter another era.

Before getting into that, I'd like to talk about some different perspectives on the kind of exponential growth we've been seeing.