[gmx-users] Splitting molecule over several CPUs

[Ute Roehrig]

Hi,

I am doing a MD simulation of several aggregated peptides in water on
4 cpus. At a certain point I decided to continue the simulation on 8
cpus. However, my prefectly stable aggregate started to rotate very
fast, and one peptide detached.

Now I ran some tests - the same simulation continued on 4 cpus (new
run input file created either with tpbconv or with grompp) shows no
rotation and no detachment. I guess this has to do with the fact that
one peptide (i.e. the one that detaches) is split over two nodes when
using 8 cpus. I am constraining only h-bonds, so grompp doesn't
complain - but is there anyways a known bug with splitting one
molecule over several nodes? If I use the "-load" option and put all
peptides on 1 cpu, everything works fine.

It's quite difficult to find this error, because at first glance the
simulation on 8 cpus seems fine, energies are OK, the rotation is not
seen if the structure is fitted, and a detachment of one peptide could
always happen...

Thanks for any suggestions,
Ute





-- 
Ute Roehrig
E. Fermi Research Centre
Department of Physics, University of Rome "Tor Vergata"
Via della Ricerca Scientifica 1, I-00133 Rome, Italy
Phone: +39-06-7259 4283, Fax: +39-06-2025 259