Here is the instructions to run NAMD on comet supercomputer:

your work directory is as follows:   /oasis/scratch/comet/username
all the files that you need to start a simulation are:
equil.scr   (for equilibrating a system with membrane)
dyn.csh
dyn.scr
last.seqno
dyn-1.inp
dyn-2.inp

These files can be accessed in the following directory: /oasis/scratch/comet/jklauda/temp_project/pla18/popc/r1/namd
 
The following is an example of using GPU nodes on comet: 
In this example we are using 4 gpu nodes of type k80. Users should always set --ntasks-per-node equal to 6 x [number of GPUs] requested on all k80  "gpu-shared" jobs, and 7 x [number of GPUs] requested on all p100 "gpu-shared" jobs, to ensure proper resource distribution by the scheduler.

#!/bin/csh

#SBATCH -A mda123

#SBATCH --job-name=popcr1

#SBATCH --partition=gpu

#SBATCH --nodes=1

#SBATCH --ntasks-per-node=24

#SBATCH --export=ALL

#SBATCH --no-requeue

#SBATCH --gres=gpu:k80:4

#SBATCH -t 24:00:00

#


More information about comet supercomputer can be accessed from the following link: https://www.sdsc.edu/support/user_guides/comet.html