Installation

ACEMD utilizes GPUs for state-of-the-art performance in molecular dynamics simulations. While it can be run on other platforms using OpenCL, for the best performance it should run on NVIDIA GPUs which use the CUDA parallel computing platform.

ACEMD can be installed either through pip or through conda, for Linux, Windows and MacOS and all current Python versions.

As conda uses an automatic CUDA detection mechanism which doesn’t always work properly, we recommend installing the correct CUDA version of ACEMD for your drivers using pip.

Detecting CUDA version

ACEMD is currently released for three CUDA versions: 11.8, 12.6 and 12.8.

To check the CUDA version of your installed NVIDIA drivers you can use the following command:

nvidia-smi

In the top right corner of the results you will see your NVIDIA driver’s CUDA version. Keep in mind that NVIDIA driver’s CUDA versions are backwards compatible, meaning that if you have a driver with CUDA 12.4, you should install the 11.8 version of ACEMD as the 12.6 version of ACEMD will not work with your drivers.

Installing from pip

To install ACEMD from pip, use the following command:

CUDA=118; pip3 install acemd --extra-index-url https://download.pytorch.org/whl/cu$CUDA --extra-index-url https://us-central1-python.pkg.dev/pypi-packages-455608/cu$CUDA/simple

You can select which CUDA version of ACEMD to install by setting the CUDA environment variable to the desired version (118, 126 or 128 corresponding to CUDA 11.8, 12.6 and 12.8 respectively).

The first --extra-index-url argument is for obtaining the correct PyTorch version which is necessary for NNP and NNP/MM simulations. The second URL is for obtaining the correct ACEMD package and its dependencies.

Installing from conda

ACEMD can also be installed with any variant of the Conda package manager. We recommend using the Miniforge installer.

After the installation of conda use the following commands to install ACEMD

conda create -n acemd -y
conda activate acemd
conda install acemd cuda-version=11.8 -c acellera -c conda-forge

If you want to install ACEMD in a pre-existing environment, just activate that environment and call the last command.

If your drivers support CUDA 12.0 or higher, you can install the CUDA 12.0 version of ACEMD by calling:

conda install acemd cuda-version=12 -c acellera -c conda-forge

Note: for commercial use, you need a license.

Base package

ACEMD also provides an acemd-base package which installs only the base dependencies required for running classical molecular dynamics simulations. This is meant to provide a minimal installation for users who do not need the additional functionalities implemented in the acemd package.

This means that it will not automatically install the libraries required for running NNP/MM or PLUMED simulations. It will however throw an error if you try to run a simulation without the required libraries and you can then choose to install those manually.

To install the acemd-base package, use the following command:

conda install acemd-base cuda-version=11.8 -c acellera -c conda-forge

If your drivers support CUDA 12.0 or higher, you can install the CUDA 12 version of ACEMD by calling:

conda install acemd-base cuda-version=12 -c acellera -c conda-forge