I remember that in university we had a test on a toy 16-bit floating point ALU. We had to walk through each of the steps in the ALU for the usual operations, showing the bit patterns at each step and keeping track of precision loss, etc. Because we had to track state and some aspects of the test were randomized, there were a few cases where identical operations on identical inputs resulted in different outputs due to the residual state.

During the next lecture, someone sticks their hand up and says, "this FPU thing is so nondeterministic, who in their right mind would use floating point for 99% of the kinds of work that we use computers for?"

The professor says, "you should think real hard about that the next time you decide to use a floating point type in your program."