reproducibility on SGI
We have a problem with reproducibility on a SGI machine.
We seem to get different results of two identical calculations
(i.e. exactly the same starting configurations, etc.). The results
initially are within the precision of the machine, but begin to
diverge from one another and this divergence continues to grow.
Has anyone had similar experiences with SGI machines?
Any suggestions (eg. recompling) on "fixing" this problem?
your comments are appreciated.
wsonnen[ AT ]alnitak.usc.edu