Amber on the cluster

Amber (Assisted Model Building with Energy Refinement) provides molecular dynamics, energy minimization, and free-energy tools for proteins, nucleic acids, lipids, and complexes. Submit simulations through Slurm.

Login nodes are for preparation

Run minimization, equilibration, and production simulations on compute nodes. Use login nodes only for input preparation, editing, and brief environment checks.

Load the environment

module avail amber

module purge
module load amber/22gpu

Check executables:

which tleap
which pmemd.cuda
pmemd.cuda -h | head

If amber/22gpu is unavailable, use an actual module from the listing or ask the administrator.

Copy the example

The public example is located at:

/hpcfs/fpublic/script_example/amber

Copy it before making changes:

cp -r /hpcfs/fpublic/script_example/amber ~/amber-demo
cd ~/amber-demo
mkdir -p output

File	Purpose
test.pdb	Initial molecular structure
tleap.in	LEaP model and force-field settings
test.prmtop	Topology and force-field parameters
test.inpcrd	Initial coordinates
md.in	Simulation controls

Generate topology and coordinates

Example tleap.in:

source leaprc.protein.ff14SB

mol = loadpdb test.pdb
check mol
saveamberparm mol test.prmtop test.inpcrd
quit

Run and inspect:

tleap -f tleap.in | tee tleap.log

ls -lh test.prmtop test.inpcrd
tail -n 20 tleap.log

The end of tleap.log should report Errors = 0. Resolve missing atoms, unknown residues, or missing parameters before simulation.

Demonstration input

This md.in is a short implicit-solvent test:

Short implicit-solvent MD test
&cntrl
  imin=0,
  irest=0,
  ntx=1,
  ntb=0,
  igb=1,
  ntt=3,
  gamma_ln=2.0,
  tempi=300.0,
  temp0=300.0,
  nstlim=50000,
  dt=0.002,
  cut=999.0,
  ntpr=500,
  ntwx=500,
  ntwr=5000,
  ioutfm=1,
/

Not a production research protocol

A research workflow normally requires structure checks, minimization, heating, appropriate density and temperature/pressure equilibration, and production simulation. Choose stages and parameters suitable for the system, solvent model, force field, and scientific question, and assess convergence and physical plausibility.

GPU submission

Save submit_amber.slurm:

#!/bin/bash
#SBATCH --job-name=amber-md
#SBATCH --partition=qgpu_3090
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --mem=16G
#SBATCH --gres=gpu:1
#SBATCH --time=02:00:00
#SBATCH --output=amber_%j.out
#SBATCH --error=amber_%j.err

set -euo pipefail

module purge
module load amber/22gpu

WORK_DIR="$HOME/amber-demo"
cd "$WORK_DIR"
mkdir -p output

echo "Job ID: ${SLURM_JOB_ID}"
echo "Node: $(hostname)"
echo "Started: $(date)"
nvidia-smi

pmemd.cuda -O \
  -i md.in \
  -o output/md.out \
  -p test.prmtop \
  -c test.inpcrd \
  -r output/md.rst7 \
  -x output/md.nc \
  -inf output/mdinfo

echo "Finished: $(date)"

Set WORK_DIR to your working directory, then submit:

sbatch submit_amber.slurm

Use pmemd.cuda for GPU jobs. A CPU-only pmemd executable will not use an allocated GPU.

Monitor and inspect results

squeue -u "$USER"

Replace 作业ID with the actual job ID:

sacct -j 作业ID --format=JobID,State,Elapsed,AllocTRES,MaxRSS,ExitCode

tail -n 50 output/md.out
tail -n 50 amber_作业ID.err

Output	Contents
output/md.out	Simulation, energy, and performance information
output/md.rst7	Final restart coordinates
output/md.nc	NetCDF trajectory
output/mdinfo	Runtime status summary

CPU parallel jobs

Request a CPU partition and use pmemd.MPI when CPU parallelism is required:

srun pmemd.MPI -O \
  -i md.in \
  -o output/md.out \
  -p test.prmtop \
  -c test.inpcrd \
  -r output/md.rst7 \
  -x output/md.nc \
  -inf output/mdinfo

Choose CPU, memory, and partition settings based on tests with your system.

Troubleshooting

If pmemd.cuda is missing, check the loaded GPU module and executable path. For LEaP residue/parameter errors, check PDB naming, ligands, termini, and force fields; nonstandard residues often require additional parameters. For CUDA initialization failures, verify --gres=gpu:1, the assigned GPU node, nvidia-smi, and logs. If no results appear, inspect the error log, output/md.out, working directory, input files, and module version.

Demonstration