To the best of my knowledge most, if not all, CFD programs in general use by aircraft manufacturers and others are indeed deterministic - the same set of initial conditions will ALWAYS produce the same solution. There is no chaos theory/random element in the workings of the program.
The 'Holy Grail' of CFD is the fully applied Navier-Stokes equations, solved throughout the flow field; it's a very very difficult problem, and for many practical cases one uses significant approximations, with empirical modelling and an understanding of the limitations of the method used to restrict use to meaningful analyses. Genghis' comment about turbulent flow is a case in point, and most CFD users understand that their tool is limited (but not all CFD advocates, some of whom could easily make a living in the marketing department one suspects).
CFD is very useful as an incremental tool, or for cases where the flow can be more simply modelled. Some conditions (such as the natural stall of a wing) are less amenable.