NAMD

Users can find the installed versions of NAMD by typing:

module spider namd

the user will be then offered the following versions of NAMD (subject to change - check with ml spider): 

NAMD/2.12-mpi
NAMD/2.12-nompi

the first option (mpi) refers to the MPI-Shared memory (SMP) version which can run on multiple nodes and can make use of worker threads on each node (process). The second option (nompi)  is the multicore version which can only run on a single node, but because it makes use of local memory it is faster than the MPI-SMP version in this case. Thus, we recommend the nompi version if you are planning to use a single node. 

Note: currently, only the nompi option supports the GPU version of NAMD in our system.  

Submit file examples

Here we present a description of our recommendation on how to run NAMD depending on the version and the number of cores you want to use:

Single node using 8 cores

#!/bin/bash
#SBATCH -A SNICXXXX-Y-ZZ
#Asking for 10 min.
#SBATCH -t 00:10:00
#Number of nodes
#SBATCH -N 1
#Ask for 8 processes
#SBATCH -c 8
#Load modules necessary for running NAMD
module add icc/2017.1.132-GCC-6.3.0-2.27  impi/2017.1.132
module add NAMD/2.12-nompi
#Execute NAMD
namd2 +setcpuaffinity +p6 config_file > output_file

Single node all cores

#!/bin/bash
#SBATCH -A SNICXXXX-Y-ZZ
#Asking for 10 min.
#SBATCH -t 00:10:00
#Number of nodes
#SBATCH -N 1
#Ask for 28 processes
#SBATCH -c 28
#Load modules necessary for running NAMD
module add icc/2017.1.132-GCC-6.3.0-2.27 impi/2017.1.132
module add NAMD/2.12-nompi
#Execute NAMD
namd2 +setcpuaffinity +p28 config_file > output_file

Using single node GPU version

#!/bin/bash
#SBATCH -A SNICXXXX-Y-ZZ
#Asking for 10 min.
#SBATCH -t 00:10:00
#Number of nodes
#SBATCH -N 1
#Ask for 28 processes
#SBATCH -c 28
#SBATCH --exclusive
#Ask for 2 GPU cards
#SBATCH --gres=gpu:k80:2
#Load modules necessary for running NAMD
module add GCC/5.4.0-2.26  CUDA/8.0.61_375.26  OpenMPI/2.0.2
module add NAMD/2.12-nompi
#Execute NAMD
namd2 +setcpuaffinity +p28 +idlepoll +devices $CUDA_VISIBLE_DEVICES config_file > output_file

Using multi-node CPU version (change as needed for a largemem node)

Here the user can execute several tasks on each node:

#!/bin/bash
#SBATCH -A SNICXXXX-Y-ZZ
#Asking for 10 min.
#SBATCH -t 00:10:00
#Number of nodes
#SBATCH -N 2
#Ask for 56 processes (2x28 cores on the nodes)
#SBATCH -n 56
#SBATCH --exclusive
#Load modules necessary for running NAMD
module add GCC/6.3.0-2.27  OpenMPI/2.0.2
module add NAMD/2.12-mpi
#Execute NAMD

srun namd2 +setcpuaffinity config_file > output_file

The option +setcpuaffinity in these examples is used to set the affinity of the working threads which results in an increase of performance > 20% w.r.t. to the timings where that option is not used. 

We evaluated the performance of the two versions of NAMD installed on Kebnekaise for a 23074 atoms system in 15,000 time steps simulation to give you an idea of the scaling behavior of this software:

We also tried a larger system to monitor the performance of NAMD.We evaluated the performance of different NAMD implementations including SMP+MPI (with 28 cores), SMP (with  28 threads), and SMP+GPU (with 2 GPU cards). The figure below shows the best performance of NAMD. The benchmark case consisted of 158944 particles, using 1 fs. for time step and a cutoff of 1.2 nm. for real space electrostatics calculations. Particle mesh Ewald was used to solve long-range electrostatic interactions. As a comparison, we used the same test case for Gromacs (see Gromacs page). 

Using multi-node GPU version

The steps that should be followed in order to use a multi-node GPU version of NAMD can be found here.