Computing Power.

First published 30 March 2016

Some 44 years ago Moore’s Law predicted a doubling of computer power every two years and it followed a very similar pattern. 222 =>4 million. It remained roughly true for so long that a Moore’s Law Law was suggested which said that the number of people predicting its demise doubled every two years. It has been suggested that this should be every 2.5 years now. The increase in power is staggering. Intel’s first commercial microprocessor chip contained 2,300 tiny transistors. The Intel Skylake contains 1.75 billion. The rate of production of transistors in 2015 was about 13 trillion per second. The distance between transistors is now so small that the wavelength of light precludes optical observation as it gets down towards atom level. From a gap between transistors of 40,000 nanometres (billionths of a metre) we are now down to 10 and at that gap internal current leakages start to occur. Future advances therefore tend to rely on new technologies, for example the D-Wave 2X quantum computer, cooled to 0.015° Kelvin and shielded from the earth’s magnetic field.  Other advances amplify this effect – specialised computing architectures and cloud based number crunchers. Progress is typified by the ability of computers to play games against humans. Until recently draughts was the best it could do to beat humans – with brute force power to optimise the 1020 possible game combinations. Chess was a different proposition because although it has 1047 possible game combinations options can be filtered as the game progresses because of the way it is structured. The latest game to be taken on is the ancient Chinese game of Go. The board has a 19×19 grid and there are approximately 250 legal moves each turn, with numerous turns involved. This gives a mind boggling 10170 possible combinations which utterly dwarfs the estimated 1080 number of particles in the observable universe. This is too high a number for brute force solving. Researchers at Alpha-Go have come up with a programme that has beaten a South Korean master of the game 4-1. The computer on which it has been run has watched million of games to extract the finer points of the principles. When it itself is playing the first algorithm uses that research to pick a handful of promising moves and a second algorithm takes those and looks at all possible combinations of moves for just a few moves ahead, comparing them to past wins. For simpler video games etc computers have learned how to play and win just by watching, with no human intervention beyond the non-specific software. The computer just looks and learns how to play. Next on the agenda – card games, just from watching games played. (The Economist.)

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s