[srimech]

That's mine. It's a physical implementation of a 5-symbol, 2-state Turing machine which was shown to be universal by Wolfram. If it were perfectly reliable (it isn't) and had an infinite tape (it doesn't) then it would be universal. Going to the extremes of low symbols and states makes it extremely inefficient in terms of code density, if you can apply that term to a Turing machine. The shortest program I've been able to write for it, doing unary subtraction of 3 and 2 would take about 60 hours and need a tape three times longer than I have.

[arethuza]

Thanks for the correction - I doubted if such a relatively simple device could be universal (see comment below), but it's fascinating to learn that it is! :-)

[Tichy]

Still, I am not sure how big other physical computers are (do any even exist)? I think it's cool :-)

Is there a compiler for it? :-)

[srimech]

It calculates a cellular automata called Rule 110 (http://en.wikipedia.org/wiki/Rule_110) which is very difficult to code for. I've seen a paper which shows a method for coding any Turing machine into rule 110, but most of it is beyond my mathematical ability at the moment. So far as I know nothing resembling a normal compiler exists for it.

It's actually much simpler to make a mechanical implementation of rule 110 directly, but Turing machines are more widely recognised.