[gmx-users] continuing a MD simulation on different architectures
Mark.Abraham at anu.edu.au
Thu Apr 2 15:13:53 CEST 2009
andrea spitaleri wrote:
> thanks for the answer. One thing: what do you mean for "but not binary identical"?
>>From the manual mdrun says:
> "The only disadvantage of dynamic load balancing is that runs are no longer binary reproducible, but
> in most cases this is not important"
> this sentence is also not very clear to me.
It means that the same inputs won't produce identical outputs, since the
load balancing means that unknowable inputs from variations in the
background work your computer system is doing will influence the timing,
and thus the balancing, and thus the 10th decimal place of the results,
and the butterfly effect ensues. You're still in the same ensemble
(within the MM approximation) but you are getting a different sample.
More information about the gromacs.org_gmx-users