It is well-known that the x86 instruction set is baroque, overcomplicated, and redundantly redundant. We show just how much fluff it has by demonstrating that it remains Turing-complete when reduced to just one instruction. The instruction we choose is mov, which can do both loads and stores. We use no unusual addressing modes, self-modifying code, or runtime code generation. Using just this instruction (and a single unconditional branch at the end of the program to make nontermination possible), we demonstrate how an arbitrary Turing machine can be simulated.

http://www.cl.cam.ac.uk/~sd601/papers/mov.pdf