Amber is developed by: David Case at Rutgers University, Tom Cheatham at the University of Utah, Ken Merz at Michigan State University, Adrian Roitberg at the University of Florida, Carlos Simmerling at SUNY-Stony Brook, Scott LeGrand at NVIDIA, Darrin York at Rutgers University, Ray Luo at UC Irvine, Junmei Wang at the University of Pittsburgh, Maria Nagan at Stony Brook, Ross Walker at GSK, and many others. Amber was originally developed under the leadership of Peter Kollman.

The following sub-tabs have examples for each version with their cooresponding dependency builds:

amber/20
amber/22
amber/22-ambertools23
amber/26 (default)

Amber version 20 is built with both CPU and GPU compatibility

Module Name: amber/20-gpu (GPU)is built with gcc 8.5.0, openmpi 4.1.3, and CUDA 11.3

Amber/20-gpu is compiled to run on a100, v100x, v100, and p100 GPU types

Module Name: amber/20.gcc (CPU) is built with gcc 8.5.0 and openmpi 4.1.3

Sample batch script for amber/20-gpu on 1 GPU

#!/bin/bash

module load amber/20-gpu

export CUDA_VISIBLE_DEVICES=0
$AMBERHOME/bin/pmemd.cuda -O -i mdin.GPU -o mdout.GPU -p prmtop -c inpcrd -r restrt.GPU -x mdcrd.GPU

For GPU jobs, you can submit with the following:

sbatch -p gpu --gres=gpu:a100:1 --ntasks-per-core=1 gpu.test

Sample MPI batch script for amber/20.gcc

#!/bin/bash

module load amber/20.gcc

mpirun -np $SLURM_NTASKS $AMBERHOME/bin/pmemd.MPI -O -i mdin.CPU -o mdout.CPU -p prmtop -c inpcrd -r restrt.CPU -x mdcrd.CPU

sbatch  --ntasks=# --ntasks-per-core=1 --nodes=1 cpu.test  #ntasks < 16
sbatch -p multinode  --ntasks=# --ntasks-per-core=1 cpu.test  #ntasks > 16

where:

-p multinode	Submit to the multinode partition where all nodes are Infiniband-connected
--ntasks #	the number of MPI processes you wish to run.
--ntasks-per-core=1	ensures that Gromacs will only run 1 MPI process per physical core (i.e will not use both hyperthreaded CPUs). This is recommended for parallel jobs.
--time=168:00:00	max walltime=168 hrs (1 week). See the section on chaining jobs below.
--exclusive	Allocate the nodes exclusively to this job (recommended for parallel jobs)

Documentation

Amber documentation for 2020

Amber version 22 is built with both CPU and GPU compatibility

Module Name: amber/22-gpu (GPU)is built with gcc 8.5.0, openmpi 4.1.3, and CUDA 11.3

Amber/22-gpu is compiled to run on a100, v100x, v100, and p100 GPU types

Module Name: amber/22.gcc (CPU) is built with gcc 8.5.0 and openmpi 4.1.3

Sample batch script for amber/22-gpu on 1 GPU

#!/bin/bash

module load amber/22-gpu

export CUDA_VISIBLE_DEVICES=0
$AMBERHOME/bin/pmemd.cuda -O -i mdin.GPU -o mdout.GPU -p prmtop -c inpcrd -r restrt.GPU -x mdcrd.GPU

For GPU jobs, you can submit with the following:

sbatch -p gpu --gres=gpu:a100:1 --ntasks-per-core=1 gpu.test

Sample MPI batch script for amber/22.gcc

#!/bin/bash

module load amber/22.gcc

mpirun -np $SLURM_NTASKS $AMBERHOME/bin/pmemd.MPI -O -i mdin.CPU -o mdout.CPU -p prmtop -c inpcrd -r restrt.CPU -x mdcrd.CPU

sbatch  --ntasks=# --ntasks-per-core=1 --nodes=1 cpu.test  #ntasks < 16
sbatch -p multinode  --ntasks=# --ntasks-per-core=1 cpu.test  #ntasks > 16

where:

-p multinode	Submit to the multinode partition where all nodes are Infiniband-connected
--ntasks #	the number of MPI processes you wish to run.
--ntasks-per-core=1	ensures that Gromacs will only run 1 MPI process per physical core (i.e will not use both hyperthreaded CPUs). This is recommended for parallel jobs.
--time=168:00:00	max walltime=168 hrs (1 week). See the section on chaining jobs below.
--exclusive	Allocate the nodes exclusively to this job (recommended for parallel jobs)

Documentation

Amber documentation for 2022

Amber version 22-ambertools23 is built with both CPU and GPU compatibility

Module Name: amber/22-ambertools23-gpu.gcc (GPU)is built with gcc 9.2.0, openmpi 4.1.4, and CUDA 11.0

Amber/22-ambertools23-gpu.gcc is compiled to run on a100, v100x, v100, and p100 GPU types

Module Name: amber/22-ambertools23.gcc (CPU) is built with gcc 9.2.0 and openmpi 4.1.4

Sample batch script for amber/amber/22-ambertools23-gpu.gcc on 1 GPU

#!/bin/bash

module load amber/22-ambertools23-gpu.gcc

export CUDA_VISIBLE_DEVICES=0
$AMBERHOME/bin/pmemd.cuda -O -i mdin.GPU -o mdout.GPU -p prmtop -c inpcrd -r restrt.GPU -x mdcrd.GPU

For GPU jobs, you can submit with the following:

sbatch -p gpu --gres=gpu:a100:1 --ntasks-per-core=1 gpu.test

Sample MPI batch script for amber/22-ambertools23.gcc

#!/bin/bash

module load amber/22-ambertools23.gcc

mpirun -np $SLURM_NTASKS $AMBERHOME/bin/pmemd.MPI -O -i mdin.CPU -o mdout.CPU -p prmtop -c inpcrd -r restrt.CPU -x mdcrd.CPU

sbatch  --ntasks=# --ntasks-per-core=1 --nodes=1 cpu.test  #ntasks < 16
sbatch -p multinode  --ntasks=# --ntasks-per-core=1 cpu.test  #ntasks > 16

where:

-p multinode	Submit to the multinode partition where all nodes are Infiniband-connected
--ntasks #	the number of MPI processes you wish to run.
--ntasks-per-core=1	ensures that Gromacs will only run 1 MPI process per physical core (i.e will not use both hyperthreaded CPUs). This is recommended for parallel jobs.
--time=168:00:00	max walltime=168 hrs (1 week). See the section on chaining jobs below.
--exclusive	Allocate the nodes exclusively to this job (recommended for parallel jobs)

Documentation

Documentation for Amber 22 and Ambertools 23

Amber version 26 is currently the default version on Biowulf

Amber version 26 is built with both CPU and GPU compatibility

Module Name: amber/26-gpu (GPU) is built with gcc 11.3.0, openmpi 4.1.7, and CUDA 12.8

Amber/26-gpu is compiled to run on a100, v100x, and v100 GPU types

Module Name: amber/26-gcc (CPU) is built with gcc 11.3.0 and openmpi 4.1.7

Sample batch script for amber/26-gpu on 1 GPU


#!/bin/bash

module load amber/26-gpu

export CUDA_VISIBLE_DEVICES=0
$AMBERHOME/bin/pmemd.cuda -O -i mdin.GPU -o mdout.GPU -p prmtop -c inpcrd -r restrt.GPU -x mdcrd.GPU

For GPU jobs, you can submit with the following:

sbatch -p gpu --gres=gpu:a100:1 --ntasks-per-core=1 gpu.test

Sample MPI batch script for amber/26-gcc

#!/bin/bash

module load amber/26-gcc

mpirun -np $SLURM_NTASKS $AMBERHOME/bin/pmemd.MPI -O -i mdin.CPU -o mdout.CPU -p prmtop -c inpcrd -r restrt.CPU -x mdcrd.CPU

sbatch  --ntasks=# --ntasks-per-core=1 --nodes=1 cpu.test  #ntasks < 16
sbatch -p multinode  --ntasks=# --ntasks-per-core=1 cpu.test  #ntasks > 16

where:

-p multinode	Submit to the multinode partition where all nodes are Infiniband-connected
--ntasks #	the number of MPI processes you wish to run.
--ntasks-per-core=1	ensures that Gromacs will only run 1 MPI process per physical core (i.e will not use both hyperthreaded CPUs). This is recommended for parallel jobs.
--time=168:00:00	max walltime=168 hrs (1 week). See the section on chaining jobs below.
--exclusive	Allocate the nodes exclusively to this job (recommended for parallel jobs)

Documentation

Amber documentation for 2026

Amber 26 is able to scale on multiple GPUs that support peer-to-peer communication (see the Amber GPU page for an explanation of peer-to-peer communication).

Please keep in mind that the performance in most cases is worse on 2 (and 4) GPUs than on a single GPU. If you plan to run a job on 2 GPUs, please run benchmarks first and verify that the performance is better on 2 GPUs than on 1. (Benchmarks). Note that the batch system will set the variable $CUDA_VISIBLE_DEVICES to the allocated GPUs.

You can check the behaviour of your job with the 'nvidia-smi' utility. Determine the GPU node on which your job is running via jobload or sjobs. Suppose your job is on node cn0626, and is using 2 GPUs:

Walltime limits are set on most Biowulf partitions. Type 'batchlim' to see the current walltime limits, or see the systems status page. Note that the default walltime on the norm queue is 4 hrs, but you can extend this to 10 days. Amber jobs should be designed to run for a week or so, save a checkpoint file, and submit a new job starting from that checkpoint.

An example batch script is below. This script runs a single simulation, saves a copy of the output files, and then resubmits a new job starting from Amber's 'restart' file.

To submit this job, copy the original input coordinate file to 'restrt' for the first run, and then submit.

Based on the benchmarks, it is highly recomended that you run Amber on a GPU node.