[AMBER] Reservoir REMD using GPUs

From: Hurst, Travis C (MU-Student) <"Hurst,>
Date: Wed, 16 May 2018 22:25:55 +0000

Hello Everyone,

I have been doing some REMD simulations on a small (20 nt) RNA hairpin in explicit solvent using Amber16, which have been going great thanks to the manual and help from the archive. I am using the results from these runs to build PMFs for the hairpin.

Now, I am interested in using a high temperature structure reservoir to speed up the convergence of the simulations. However, when I try to use pmemd.cuda.MPI to run the simulations, pmemd does not recognize the -rremd flag. Is there a way to use GPUs for running R-REMD simulations or is sander.MPI the only option? I have gotten the R-REMD simulation to work using sander.MPI, but mdinfo estimated about 5700 hours to finish, so incorporating the GPUs is definitely necessary for production runs if I use this method.

Additionally, I found the createreservoir command in the manual and built a structure reservoir using that, but I am not sure how to load this into memory for the R-REMD simulations to properly read the files. I tried placing the name of the reservoir after the -reservoir flag, but the simulation cannot find the files. Before finding the createreservoir command, I built a reservoir by opening cpptraj, reading in the parm/trajectory files, and using trajout ./reserv/frame restart to build the reservoir. Then, I changed the file numbers from the command line to fit the format from the manual. This seems to work ok, but the createreservoir command works faster and uses much less memory.

Finally, if anyone has any advice on other enhanced sampling methods that may be more appropriate for rapidly getting accurate PMFs for this system, I am all ears.


AMBER mailing list
Received on Wed May 16 2018 - 15:30:02 PDT
Custom Search