How well can we predict our future climate? If the flap of a butterfly’s wings can change the course of weather a week or so from now, what hope trying to predict anything about our climate a hundred years hence? In this talk I will discuss the science of climate change from a perspective which emphasises the chaotic (and hence uncertain) nature of our climate system. In so doing I will outline the fundamentals of climate modelling, and discuss the emerging concept of inexact supercomputing, needed - paradoxically perhaps - if we are to increase the accuracy of predictions from these models. Indeed, revising the notion of a supercomputer from its traditional role as a fast but precise deterministic calculating machine, may be important not only for climate prediction, but also for other areas of science such as astrophysics, cosmology and neuroscience.