How does 'validation' work, why does it sometimes fail?

Hi!

I am asking a question that may well have already been asked, if so, just show me the 'exit' but perhaps in the form of some kind of references to earlier articles, for which I have searched.

Are we dealing with deterministic processes here? If we are, how could two runs on the same data set fail to produce the identical result?

Is a failure of validation perhaps due to differing implementations of floating point processors in different CPU chips?

I don't like seeing my machines grind away for 'nothing' (so to speak) but my reason for asking is to comprehend the problem of validation.

What's up?

Thanks and all the best!