Years ago, when we all had single processor PC's, chess engines were determinisitc in their play and evaluations.
It's different these days. There's a bit of randomness. Maybe this thread at Open Chess is what you're looking for ---->> Deterministic chess algorithm
Komodo 8 often gives different evaluations for moves that shortly transpose into the same variation. I've seen a difference in evals in this situation up to a tenth of a pawn. This isn't a major problem, but seems odd.
I believe I've also seen the evaluations differ for the same position at the same ply depth if the position is run again on a different day.
I can't really explain the second situation. Why does this happen?