Why do I get slightly different answers on different computers?
An apparently identical run on a different computer will usually yield different results within statistical fluctuations since the floating point representations among different architectures is different. Imagine that a simulation takes a branch at a point, testing whether a random number, RN, is less than, say 0.5. Architecture 1 may produce a random number of 0.4999999 and architecture 2 0.5000001. The different machines then take different branches and the simulations are different.
