Moore power

Some things in this world never cease to amaze me. I still remember as a teenager reading about a super computer being built that would perform 40 billion calculations per second. The thought of what could be done with so much computation power was nothing less than inspiring… Since then, I’ve been keeping an eye out on the power of the largest super computers in the world. Every time a new machine sets a seemingly unsurpassable performance record, within a few years another machine comes along to break it, usually by a factor of 3 or more.
Earlier this year IBM’s Blue Gene/L super computer set a new world record with 2.8 x 10¹⁴ FLOPS. A few months ago another group reported that their machine had managed 10¹⁵ FLOPS. Then yesterday I read in the BBC that IBM’s next machine, called Roadrunner, should be capable of 1.6 x 10¹⁵ FLOPS when completed in 2008. Plotting all this on a graph:

I asked Open Office to fit a curve to this data, and if you look carefully you’ll see that the growth is slightly super exponential — something that Ray Kurzweil likes to make a point of. Kurzweil looks like he might be wrong about one thing however: He was predicting that super computer performance would reach 10¹⁶ FLOPS in about 2013. For once he might have been a little too conservative.

What we need now is someone to come up with a way to turn ridiculous amounts of computation into real artificial intelligence.