It does not have to use floating point numbers, while floating point numbers are hardware accelerated and thus very fast, there are infinite precision number libraries.

But more likely, the program would have to be built where it understands the symbolic forms involved, more like a proof solver than simple cfd math.