Different Bayesian estimation results between Windows and IOS

Hi everyone, I find that when I use Bayesian estimation, the post mean of parameters may be a little different in Windows and IOS environments. I both use dynare 5.0, what’s problems with that?

How different are we talking about?

For example, the post mean of the degree of habit persistence(gam) in Windows is 0.8634, while it is 0.8631 in IOS environment. the post mean of the persistence of the housing demand shock(rho_phip) in Windows is 0.9864, while it is 0.9860 in IOS.
There is little difference between two environment, but impulse respinses looks like the same.

My hunch is that small differences either arise due to differences in random number generators or matrix kernel libraries.