AMBER

Introduction¶

Amber is the collective name for a suite of programs that allow users to perform molecular dynamics simulations, particularly on biomolecules. None of the individual programs carry this name, but the various parts work reasonably well together and provide a powerful framework for many common calculations.

Amber modules¶

We provide modules for Amber, AmberTools, and Amber-PMEMD in our software stack.

AmberTools (module ambertools) - Tools for preparing/analyzing simulations, QUICK for GPU-accelerated DFT calculations and sander for molecular dynamics. Free and open source.
Amber (module amber) - Everything included in AmberTools, plus the advanced pmemd program for high-performance molecular dynamics simulations.
Amber-PMEMD (module amber-pmemd, Amber 24+) – High-performance MD engine pmemd, optimized for CPU and GPU. Provides the high-performance MD engine pmemd (optimized for CPU/GPU) as a standalone module. This change was made because starting with Amber 24, pmemd no longer requires AmberTools for compilation.

Amber-PMEMD Module

The amber-pmemd module does not include AmberTools. To use both, load the ambertools module as well.

To see a list of installed versions and which other modules they depend on, you can use the module spider command or check the Available software page.

Using AMBER on H100 GPU Clusters¶

H100 GPU Compatibility Update

Older AMBER modules are incompatible with NVIDIA H100 GPUs. For GPU-accelerated runs, use the newly installed modules below.

Module Requirements:¶

ambertools/25.0 or amber-pmemd/24.3

These modules include H100-specific CUDA kernels (compiled with CUDA 12+ for the Hopper architecture).

Avoid Legacy Modules on H100

Do not use legacy AMBER modules for GPU jobs — they will fail on H100 nodes.

Loading modules¶

StdEnv/2023

AMBER version	modules for running on CPUs	modules for running on GPUs (CUDA)	Notes
`amber-pmemd/24.3`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3`	H100 compatible
`amber/22.5-23.5`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 amber/22.5-23.5`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.2 amber/22.5-23.5`
`ambertools/25.0`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 ambertools/25.0`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 ambertools/25.0`	H100 compatible, with PLUMED/2.9.0
`ambertools/23.5`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 ambertools/23.5`	`StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.2 ambertools/23.5`

StdEnv/2020

AMBER version	modules for running on CPUs	modules for running on GPUs (CUDA)	Notes
`ambertools/21`	`StdEnv/2020 gcc/9.3.0 openmpi/4.0.3 scipy-stack ambertools/21`	`StdEnv/2020 gcc/9.3.0 cuda/11.4 openmpi/4.0.3 scipy-stack ambertools/21`	GCC, FlexiBLAS & FFTW
`amber/20.12-20.15`	`StdEnv/2020 gcc/9.3.0 openmpi/4.0.3 amber/20.12-20.15`	`StdEnv/2020 gcc/9.3.0 cuda/11.4 openmpi/4.0.3 amber/20.12-20.15`	GCC, FlexiBLAS & FFTW
`amber/20.9-20.15`	`StdEnv/2020 gcc/9.3.0 openmpi/4.0.3 amber/20.9-20.15`	`StdEnv/2020 gcc/9.3.0 cuda/11.0 openmpi/4.0.3 amber/20.9-20.15`	GCC, MKL & FFTW
`amber/18.14-18.17`	`StdEnv/2020 gcc/9.3.0 openmpi/4.0.3 amber/18.14-18.17`	`StdEnv/2020 gcc/8.4.0 cuda/10.2 openmpi/4.0.3`	GCC, MKL

StdEnv/2016

AMBER version	modules for running on CPUs	modules for running on GPUs (CUDA)	Notes
`amber/18`	`StdEnv/2016 gcc/5.4.0 openmpi/2.1.1 scipy-stack/2019a amber/18`	`StdEnv/2016 gcc/5.4.0 openmpi/2.1.1 cuda/9.0.176 scipy-stack/2019a amber/18`	GCC, MKL
`amber/18.10-18.11`	`StdEnv/2016 gcc/5.4.0 openmpi/2.1.1 scipy-stack/2019a amber/18.10-18.11`	`StdEnv/2016 gcc/5.4.0 openmpi/2.1.1 cuda/9.0.176 scipy-stack/2019a amber/18.10-18.11`	GCC, MKL
`amber/18.10-18.11`	`StdEnv/2016 gcc/7.3.0 openmpi/3.1.2 scipy-stack/2019a amber/18.10-18.11`	`StdEnv/2016 gcc/7.3.0 cuda/9.2.148 openmpi/3.1.2 scipy-stack/2019a amber/18.10-18.11`	GCC, MKL
`amber/16`	`StdEnv/2016.4 amber/16`		Available only on Graham. Some Python functionality is not supported

Using modules¶

AmberTools 21¶

Currently, AmberTools 21 module is available on all clusters. AmberTools provide the following MD engines: sander, sander.LES, sander.LES.MPI, sander.MPI, sander.OMP, sander.quick.cuda, and sander.quick.cuda.MPI. After loading the module set AMBER environment variables:

source $EBROOTAMBERTOOLS/amber.sh

Amber 20¶

There are two versions of amber/20 modules: 20.9-20.15 and 20.12-20.15. The first one uses MKL and cuda/11.0, while the second uses FlexiBLAS and cuda/11.4. MKL libraries do not perform well on AMD CPU, and FlexiBLAS solves this problem. It detects CPU type and uses libraries optimized for the hardware. cuda/11.4 is required for running simulations on A100 GPUs installed on Narval.

CPU-only modules provide all MD programs available in AmberTools/20 plus pmemd (serial) and pmemd.MPI (parallel). GPU modules add pmemd.cuda (single GPU), and pmemd.cuda.MPI (multi - GPU).

Known issues¶

Module amber/20.12-20.15 does not have MMPBSA.py.MPI executable.
MMPBSA.py from amber/18-10-18.11 and amber/18.14-18.17 modules cannot perform PB calculations. Use more recent amber/20 modules for this type of calculations.

Job submission examples¶

Single GPU job¶

For GPU-accelerated simulations on Narval, use amber/20.12-20.15. Modules compiled with CUDA version < 11.4 do not work on A100 GPUs. Below is an example submission script for a single-GPU job.

pmemd_cuda_job.sh

#!/bin/bash
#SBATCH --ntasks=1 
#SBATCH --gpus-per-node=1 
#SBATCH --mem-per-cpu=2000 
#SBATCH --time=10:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

pmemd.cuda -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

CPU-only parallel MPI job¶

NarvalRorqualFirNibiTrillium

pmemd_MPI_job_narval.sh

#!/bin/bash
#SBATCH --nodes=4
#SBATCH --ntasks-per-node=64
#SBATCH --mem-per-cpu=2000
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

srun pmemd.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

pmemd_MPI_job_rorqual.sh

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=192
#SBATCH --mem-per-cpu=2000
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

srun pmemd.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

pmemd_MPI_job_fir.sh

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=192
#SBATCH --mem-per-cpu=2000
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

srun pmemd.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

pmemd_MPI_job_nibi.sh

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=192
#SBATCH --mem-per-cpu=2000
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

srun pmemd.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

pmemd_MPI_job_trillium.sh

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=192
#SBATCH --mem-per-cpu=2000
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 amber-pmemd/24.3

srun pmemd.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

QM/MM distributed multi-GPU job¶

The example below requests eight GPUs.

quick_MPI_job.sh

#!/bin/bash
#SBATCH --ntasks=8
#SBATCH --gpus-per-task=1 
#SBATCH --mem-per-cpu=4000 
#SBATCH --time=02:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 ambertools/25.0

srun sander.quick.cuda.MPI -O -i input.in -p topol.parm7 -c coord.rst7 -o output.mdout -r restart.rst7

Parallel MMPBSA job¶

The example below uses 32 MPI processes. MMPBSA scales linearly because each trajectory frame is processed independently.

mmpbsa_job.sh

#!/bin/bash
#SBATCH --ntasks=32 
#SBATCH --mem-per-cpu=4000 
#SBATCH --time=1:00:00

module purge
module load StdEnv/2023 gcc/12.3 openmpi/4.1.5 cuda/12.6 ambertools/25.0

srun MMPBSA.py.MPI -O -i mmpbsa.in -o mmpbsa.dat -sp solvated_complex.parm7 -cp complex.parm7 -rp receptor.parm7 -lp ligand.parm7 -y trajectory.nc

You can modify scripts to fit your simulation requirements for computing resources. See Running jobs for more details.

Performance and benchmarking¶

A team at ACENET has created a Molecular Dynamics Performance Guide for Alliance clusters. It can help you determine optimal conditions for AMBER, GROMACS, NAMD, and OpenMM jobs. The present section focuses on AMBER performance.

View benchmarks of simulations with PMEMD

View benchmarks of QM/MM simulations with SANDER.QUICK.