I ran exactly same scripts on different machines (windows and linux). Windows has one of the 2016 versions of LAMMPS and Linux has the latest version. Simulations are TIP5P pure water models with Green-Kubo algorithm (EMD). They have similar results when I run TIP3P models, but with TIP5P, Linux machine overestimates the experimental result, while Windows' results are in good agreement with the experiment. Both machines are operating properly and have no observable problems (We use them a lot in our research and this is the only type of simulation they disagree with each other).
Can somebody explain this situation ?