Thursday, July 17, 2003

Addicted to Moore's Law

Every year, engineers at Intel and AMD deliver another nice heaping pile of CPU crack to software developers, who gratefully smoke it and produce increasingly bloated, crappy, huge, slow programs. And they get away with it, of course, because if your program gets twice as crappy every eighteen months, no one notices except the unAmerican cheapskates who refuse to buy a new computer every couple of years. Since 1995, CPUs have gotten something like 50 times faster. How much faster is Windows XP than Windows 95? Umm... not at all? And recall that Win95 ran reasonably on computers with 32 MB of memory, itself a stunning increase over the 4 and 8 MB recommendations for Windows 3. I just had to add another 256 MB of memory to my Dell laptop to stop XP from swapping every time I moved the mouse. (Okay, granted I was moving the mouse over OpenOffice. But still.)

And of course the reason that we don't storm the Redmond Bastille is that that 256 MB upgrade cost me a total of $44. (Does anyone need an extra 64 MB SODIMM?) So programmers -- and it's hardly just Microsoft -- get away with doubling the memory footprint of their programs every generation, get away with grindingly slow implementations, all saved by Gordon's magic wand.

The first question I ask myself is "Why isn't software getting better?" Are we really, really doing so much more it? My web browser is only marginally more functional than Netscape 2 was in those days; my spreadsheet calculates the same functions, my email is the same, my word processor is the same except for those squiggly red lines.

The second question I ask myself is "What happens if Moore's Law is repealed?" We all know that it isn't a law (although irritatingly the press often refers to it as if it were a physical principle), just an output of economics and engineering. The engineering won't stop, but the economics very well might. It seems to me that there's a huge market opportunity for a classically Clay Christensen disruptive innovation, undercutting price and performance of existing processors signficantly. AMD tried this, a little, and that's the theory behind the Intel Celeron, but these are on the same escalator, just a few steps behind. There's a lot of people out there who would pay a few hundred dollars for a fully functional, 1999-era computer, and that computer will be reasonably powered for many, many years. Other than a few niche applications (most notably 3D games), we've really saturated the performance market.

My prediction is that Moore's Law will end because of market economics long before it ends because of engineering limitations, and when we do, it's going to be a long, hard, cold turkey for some software companies.