Installing Amber on Various Platforms

Patricio Barletta¹, Zeke A. Piskulich¹, and Darrin M.

York¹

¹Laboratory for Biomolecular Simulation Research, Institute for Quantitative Biomedicine and Department of Chemistry and Chemical Biology, Rutgers University, Piscataway, NJ 08854, USA

Learning objectives

• Learn how to install amber on various supercomputing platforms

• Learn how to make that amber version compatible with FE-Toolkit and AmberFlow

Tutorial

Amber is one of the programs most used, and developed, by our group. Thus, you will likely be installing and using Amber on a wide range of supercomputing environments, each which requires different installation steps.

Setting Up Your Environment (All Platforms)

Before installing amber, you will need to setup your environment, specifically python.

1. Get Miniforge (a minimal conda installer) from here <https://github.com/conda-forge/miniforge>.

Install it with the following (accept all defaults):

bash Miniforge3-MacOSX-arm64.sh  # for Mac M1
bash Miniforge3-MacOSX-x86_64.sh  # for Mac Intel
bash Miniforge3-Linux-aarch64.sh  # for Linux ARM
bash Miniforge3-Linux-x86_64.sh  # for Linux Intel/AMD

2. Create a mamba environment for amber:

mamba create -n amber "python>3.11" "numpy<2" scipy matplotlib cython rdkit pyyaml -y
mamba activate amber

3. Clone the amber git repo (you must have access to it).

git clone git@gitlab.ambermd.org:amber/amber.git
mkdir debug_lbsr_dev
cd amber
git switch lbsr_dev
cd ../debug_lbsr_dev

Note

The above uses the lbsr_dev branch; however, you can use any branch you want. Usually, the lbsr_dev branch is the most up-to-date and stable for our purposes.

Installing Amber on Frontera

1. Get an interactive node

idev -p flex -t 4:00:00 -N 1

2. Load the required modules

module load gcc/9.1.0 cuda/12.2 cmake/3.24.2 impi/19.0.9

Note

While not explicitly required for amber, we recommend also loading the following modules for fe-toolkit compatibility.

module load mkl/19.1.1

3. Run CMAKE to configure the build.

cmake ../amber -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=../install_lbsr_dev -DCOMPILER=GNU -DMPI=TRUE -DCUDA=TRUE -DINSTALL_TESTS=TRUE -DDOWNLOAD_MINICONDA=FALSE -DBUILD_PYTHON=ON -DDISABLE_PBSA_CUDA=ON

4. Build and install (with parallel make)

make -j 56 install

Note

Installing FE-Toolkit on Frontera

If you want to install FE-Toolkit, use the following commands.

cd ../
git clone git@gitlab.com:RutgersLBSR/fe-toolkit.git
cd fe-toolkit
mkdir debug
cd debug
cmake .. -DCMAKE_INSTALL_PREFIX=../local -DBUILD_PYTHON=TRUE -DPython3_EXECUTABLE=`which python3` -DCMAKE_CXX_COMPILER=`which g++` -DCMAKE_Fortran_COMPILER=`which gfortran` -DCMAKE_CXX_FLAGS="-O3 -DNDEBUG -Wall -Wextra -Wunused -march=native -mtune=native"
make -j 56 install

5. Now add the following function to your .bashrc file (or an independent script that you can source when needed).

function load_amber() {

  module load gcc/9.1.0 cuda/12.2 cmake/3.24.2 impi/19.0.9 mkl/19.1.1
  mamba activate amber

  export BACKUP_PATH=$PATH
  export BACKUP_PYTHONPATH=$PYTHONPATH
  source <path to your amber install>/install_lbsr_dev/amber.sh
  export PATH=$BACKUP_PATH:$PATH
  export PYTHONPATH="<path to your amber install>/fe-toolkit/local/lib/python3.11/site-packages"
  export PATH="<path to your fe-toolkit dir>/fe-toolkit/debug/bin:$PATH"

}

6. To install amberflow (only if you have access to it).

cd ../
git clone git@gitlab.com:RutgersLBSR/amberflow.git
cd amberflow

open the pyproject.toml file and remove edgembar, since we built it from source.

Then run:

pip install -e .

You should have a working amber installation now!

Installing Amber on Vista

1. Get and interactive node

idev -N 1 -p gg -t 4:00:00

2. Load the required modules

module load nvidia/24.7 openmpi/5.0.3 cuda/12.4  nvidia_math/12.4 nccl/12.4 gcc/13.2.0 cmake/3.31.5

Note

While not explicitly required for amber, we recommend also loading the following modules for fe-toolkit compatibility.

module load nvpl

3. Run CMAKE to configure the build.

cmake ../amber -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=../install_lbsr_dev -DCOMPILER=GNU -DMPI=TRUE -DCUDA=TRUE -DNVIDIA_MATH_LIBS=${TACC_NVIDIA_MATH_LIB} -DINSTALL_TESTS=TRUE -DDOWNLOAD_MINICONDA=FALSE -DBUILD_PYTHON=ON -DDISABLE_PBSA_CUDA=ON

4. Build and install (with parallel make)

make -j 56 install

Note

Installing FE-Toolkit on Vista

If you want to install FE-Toolkit, use the following commands.

cd ../
git clone git@gitlab.com:RutgersLBSR/fe-toolkit.git
cd fe-toolkit
mkdir debug
cd debug
cmake .. -DCMAKE_INSTALL_PREFIX=../local -DBUILD_PYTHON=TRUE -DPython3_EXECUTABLE=`which python3` -DCMAKE_CXX_COMPILER=`which g++` -DCMAKE_Fortran_COMPILER=`which gfortran` -DCMAKE_CXX_FLAGS="-O3 -DNDEBUG -Wall -Wextra -Wunused -march=native -mtune=native"
make -j 56 install

5. Now add the following function to your .bashrc file (or an independent script that you can source when needed).

function load_amber() {

  module load nvidia/24.7 openmpi/5.0.3 cuda/12.4  nvidia_math/12.4 nccl/12.4 gcc/13.2.0 cmake/3.31.5 nvpl
  mamba activate amber

  export BACKUP_PATH=$PATH
  export BACKUP_PYTHONPATH=$PYTHONPATH
  source <path to your amber install>/install_lbsr_dev/amber.sh
  export PATH=$BACKUP_PATH:$PATH
  export PYTHONPATH="<path to your amber install>/fe-toolkit/local/lib/python3.11/site-packages"
  export PATH="<path to your fe-toolkit dir>/fe-toolkit/debug/bin:$PATH"

}

6. To install amberflow (only if you have access to it).

cd ../
git clone git@gitlab.com:RutgersLBSR/amberflow.git
cd amberflow

open the pyproject.toml file and remove edgembar, since we built it from source.

Then run:

pip install -e .

Installing Amber on Amarel