On Fri, Sep 14, 2018, Asier Zorrilla wrote:
>
> Our users have test a data set with Amber (same version Amber and
> Amber_tools) in two machines but they obtain two diferent results.
>
> - There is any way (beyond benchmark test) to verify that the
> installation is ok and that the obtained values are fiable?
You should run "make test" in the $AMBERHOME directory (if you have not
already done so). This tests correctness of the installation in ways
that the benchmarks do no.
>
> - Tested two systems have a GPU, both of them certified by you (GTX
> 1080 and GTX TITAN), is posible that the diferent results can be
> explained with this?
We would need to know whether the "different results" only show up on
GPU runs, or on other types of tests. If you are running the (default)
pmemd.cuda (which uses the SPFP precision model), then it is likely that
GTX 1080 and GTX Titan will give different results.) More details are
available here:
http://ambermd.org/GPULogistics.php#Accuracy
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 14 2018 - 06:00:06 PDT
