Software¶

NU HPC software is exposed primarily through Lmod environment modules. Exact versions differ by cluster and change over time, so examples use placeholders and discovery commands instead of treating the 2026 snapshot as a live catalogue.

Find and load modules¶

module purge
module spider python
module spider Python/<version>
module load Python/<version-and-toolchain>
module list

In a hierarchical module tree, module avail shows modules loadable in the current environment, while module spider searches the full known tree and explains prerequisites.

Command	Purpose
`module avail`	show modules loadable now
`module spider <name>`	search all known modules and versions
`module keyword <term>`	search module names/descriptions
`module show <module>`	inspect environment changes without loading
`module load <module>`	load a module and its dependencies
`module list`	record the active environment
`module unload <module>`	unload one module
`module purge`	return to a clean module environment

See the official Lmod user guide and module spider guide.

Make jobs reproducible

Load explicit versions in production scripts, print module list, and keep the compiler/MPI/CUDA module used at runtime consistent with the one used to build the executable.

Python¶

Lightweight virtual environment¶

Prefer a normal Python module plus venv when Conda is unnecessary:

module purge
module load <python-module>
python -m venv "$HOME/.venvs/myproject"
source "$HOME/.venvs/myproject/bin/activate"
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python -m pip freeze > requirements.lock.txt

Create environments on a shared filesystem so compute nodes can see them. Avoid installing packages from login-node source builds that consume substantial CPU or memory; use a short compute allocation for heavy compilation.

Conda environments¶

module purge
module load <conda-or-miniforge-module>
eval "$(conda shell.bash hook)"
conda create --name myproject python=3.12
conda activate myproject
conda install --name myproject numpy scipy
conda env export --from-history > environment.yml

The old page used source activate and conda --name ... install; current Conda syntax is conda activate and conda install --name .... See the official environment and package guides.

For reproducibility, prefer one well-defined channel policy and commit environment.yml. Do not place secrets for private channels in the repository.

GCC and GNU Fortran¶

module purge
module spider GCC
module load GCC
gcc -O2 -Wall -Wextra hello.c -o hello
gfortran -O2 -Wall -Wextra hello.f90 -o hello-fortran

Use -fopenmp for OpenMP code. Treat -ffast-math as an algorithmic change: it relaxes IEEE behavior and can make numerical results less reproducible. Avoid -march=native for binaries that may run on a different CPU generation; build for the oldest target architecture or use per-system builds.

MPI toolchains¶

Use wrapper compilers from a single module stack:

module purge
module spider foss
module load foss
mpicc -O2 program.c -o program
mpifort -O2 program.f90 -o program-fortran

Record mpicc --showme or the equivalent wrapper output when diagnosing linkage. Launch through the site-supported Slurm/MPI integration described in Job submission.

CUDA¶

CUDA workloads require three compatible layers: NVIDIA driver, CUDA toolkit/runtime and the application.

module purge
module spider CUDA
module load CUDA
nvcc --version

Only inspect GPUs inside an allocation:

nvidia-smi -L
nvidia-smi

The “CUDA Version” displayed by nvidia-smi is the newest CUDA API level supported by the driver, not necessarily the toolkit loaded with Lmod. Use nvcc --version and module list for the toolkit.

Gaussian¶

Gaussian is licensed software; access and available module names depend on NU licensing. Match Gaussian resources to the input file:

input.com

%NProcShared=8
%Mem=16GB
%Chk=input.chk
#p HF/6-31G(d) Opt

Water optimization

0 1
O   0.000000   0.000000   0.117790
H   0.000000   0.755453  -0.471161
H   0.000000  -0.755453  -0.471161

gaussian.slurm

#!/bin/bash
#SBATCH --job-name=gaussian
#SBATCH --partition=<cpu-partition>
#SBATCH --time=1-00:00:00
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=8
#SBATCH --mem=16G
#SBATCH --output=logs/%x-%j.out

set -euo pipefail
module purge
module load <gaussian-module>

export GAUSS_SCRDIR="${SLURM_TMPDIR:-${TMPDIR:-$HOME/scratch/$SLURM_JOB_ID}}"
mkdir -p "$GAUSS_SCRDIR"
g16 < input.com > input.log

Prefer scheduler-provided node-local scratch when available. If the fallback under $HOME is used, monitor quota and remove disposable scratch files only after confirming that results/checkpoints have been copied safely. Do not target an individual idle node; let Slurm place the job. GPU use in Gaussian is method- and build-specific and must not be advertised as a generic partition switch.

MATLAB¶

Use non-interactive batch mode for production:

matlab.slurm

#!/bin/bash
#SBATCH --job-name=matlab
#SBATCH --partition=<cpu-partition>
#SBATCH --time=01:00:00
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --mem=8G
#SBATCH --output=logs/%x-%j.out

set -euo pipefail
module purge
module load <matlab-module>
matlab -batch "run('analysis.m')"

Toolbox and license availability is site-specific. Do not hard-code the license server in user documentation.

LAMMPS¶

The module should provide its compatible compiler, MPI and accelerator dependencies. Loading unrelated GCC/OpenMPI versions afterward can break the environment.

lammps.slurm

#!/bin/bash
#SBATCH --job-name=lammps
#SBATCH --partition=<cpu-partition>
#SBATCH --time=01:00:00
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=8
#SBATCH --cpus-per-task=1
#SBATCH --mem-per-cpu=2G
#SBATCH --output=logs/%x-%j.out

set -euo pipefail
module purge
module load <lammps-module>
module list
srun lmp -in input.in

Requesting shared software¶

Request a shared installation when the software is licensed for NU, compatible with the operating system, useful to more than one workflow and able to use the cluster meaningfully. Include:

name, version and authoritative download URL;
license/access requirements;
target cluster and CPU/GPU/MPI requirements;
an installation recipe or EasyBuild easyconfig if available;
a minimal validation input and expected result;
research groups expected to use it.

User-space installations remain subject to storage quotas and security policy.