Hub

GROMACS

Molecular dynamics simulation package for biomolecular systems

GROMACS is a versatile, high-performance package for molecular dynamics simulations of proteins, lipids, and nucleic acids, backed by remote compute. This image is built once with both MPI and CUDA support: gmx_mpi uses an attached NVIDIA GPU when present and falls back to CPU/MPI when not — so the same image covers both modes.

Usage

1. Deploy

dxflow workflow create --identity gromacs hub://gromacs

# With a GPU (default), or CPU-only by dropping the gpu resource
dxflow workflow start gromacs

The container stays up so you can run gmx_mpi commands against data mounted at /volume.

2. Run a simulation

The gmx_mpi commands below run inside the workflow container; put your inputs under /volume.

# Prepare the system
gmx_mpi pdb2gmx -f protein.pdb -o processed.gro -water spce
gmx_mpi editconf -f processed.gro -o newbox.gro -c -d 1.0 -bt cubic
gmx_mpi solvate -cp newbox.gro -cs spc216.gro -o solv.gro -p topol.top

# Energy minimization
gmx_mpi grompp -f em.mdp -c solv.gro -p topol.top -o em.tpr
gmx_mpi mdrun -v -deffnm em

# Production MD (add -nb gpu on a GPU node)
gmx_mpi grompp -f md.mdp -c npt.gro -t npt.cpt -p topol.top -o md.tpr
gmx_mpi mdrun -v -deffnm md -nb gpu

3. Retrieve results

Everything under /volume persists — trajectories, logs, and analysis outputs are written there.

Configuration

Attach a GPU with resources.gpu: nvidia for CUDA acceleration, or remove it to run CPU/MPI-only.

name: gromacs
tags:
    - molecular
steps:
    - name: app
      runtime: docker
      mode: parallel
      image: ghcr.io/dxflow-ai/gromacs:latest
      command:
          - tail
          - -f
          - /dev/null
      volumes:
          - name: volume
            host: ./volume
            container: /volume
      resources:
          cpu: "4"
          memory: 32G
          gpu: nvidia

[volume]
app.volume = ./volume

[resource]
app.cpu = 4
app.memory = 32G
app.gpu = nvidia

{
    "arch": ["amd64"],
    "image": "ghcr.io/dxflow-ai/gromacs:latest",
    "version": "2025.2",
    "minimum": {
        "cpu": 2,
        "memory": "16G",
        "storage": "50G"
    }
}

Notes

  • GPU vs CPU: with a GPU attached, offload work with mdrun -nb gpu (and -pme gpu, -bonded gpu); check nvidia-smi. Without a GPU, gmx_mpi runs on CPU automatically.
  • MPI parallelism: launch multiple ranks with mpirun -np <N> gmx_mpi mdrun -v -deffnm md; use -ntomp for OpenMP threads per rank and -tunepme yes for automatic PME tuning.
  • Built from source with -DGMX_SIMD=AVX2_256 and its own bundled FFTW; the GMXRC environment is sourced for interactive shells.
  • Supports the common force fields (AMBER, CHARMM, GROMOS, OPLS), free-energy and umbrella-sampling methods, and REMD.

References

Molecular

Pre-configured workflows for simulating molecular behavior and interactions

PyMOL

Molecular visualization of structures and trajectories