Stephen M. Dutz
smdutz at csupomona.edu
Wed Jan 2 19:48:50 CET 2008
Thanks for the reply. I accidentally had lincs-iter set to 4 initially because I forgot to change it to 1 after minimization. After looking into the issue some more I determined the odd results to be due to a few stray processes running on the node I was performing the calculation on. After taking care of these stray processes the calculation time returned to an understandable 3.8 hrs/ns.
> Will someone please explain to me why I am seeing the following... I am running a protein/ lipid system with position restraints on the protein. I run two identical runs except in one I have lincs-iter = 1 and the other I have lincs-iter = 4. With lincs-iter = 1 I observe 26.0 hrs/ns computational time. With lincs-iter = 4 I observe 4.0 hrs/ns. This seems a bit backwards to me, but perhaps I have misunderstood something.
Over how long were you measuring these performance times? You'll need a
few picoseconds to be sure you're not being dominated by setup time.
> Other pertinent settings are; 2fs timestep, pbc = xyz, PME, berendsen thermostat and barostat @ 310K and 1 bar, constraints on all-bonds, Lincs constraint algorithm, lincs-order = 4
Use gmxcheck to compare your .tpr files to verify that the differences
you think exist are all that does exist.
Since you're not doing NVE, why do you want lincs_iter > 1?
"You must be the change you want to see in the world."
More information about the gromacs.org_gmx-users