IEEE floating point indeterminism means you have to use arbitrary precision math, e.g. fixed point arithmetic. So every library has to be re-written. Upshot? works great on normal CPUs, since they're the kings of integer math.