GPU-Based Installation of LAMMPS¶

1. Requirements¶

To run LAMMPS on GPU nodes, the following components are required:

Component	Purpose
NVIDIA GPU	Hardware accelerator used for high-performance simulations
CUDA Toolkit	Provides libraries and tools for GPU computing
MPI	Enables parallel execution across multiple CPU cores
CMake	Build system used to configure and compile LAMMPS
GCC / Compiler	Compiles the LAMMPS source code

LAMMPS uses GPU acceleration to offload computationally intensive operations such as pairwise force calculations between atoms, significantly improving simulation performance.

2. Check GPU Availability¶

Before installing LAMMPS with GPU support, verify that the GPU is available on the node.

nvidia-smi

Example output:

NVIDIA-SMI 535.xx
GPU Name: Tesla V100 / A100 / RTX

Purpose:

Confirms that the GPU driver is installed
Verifies that the GPU node is accessible

3. Install CUDA Toolkit¶

LAMMPS GPU support requires the CUDA Toolkit.

Check CUDA installation:

nvcc --version

If CUDA is not installed, install it:

sudo dnf install cuda

or install it from the official NVIDIA CUDA repository.

CUDA provides:

GPU libraries
CUDA compiler (nvcc)
GPU runtime environment

4. Install Required Build Tools¶

Install the necessary development tools:

sudo dnf install gcc gcc-c++ cmake git openmpi openmpi-devel

Explanation:

Tool	Purpose
gcc	Compiles C++ source code
cmake	Configures the build environment
openmpi	Enables MPI parallel execution
git	Downloads the LAMMPS source code

5. Download LAMMPS Source Code¶

LAMMPS source code can be downloaded from the official website:

https://www.lammps.org/download.html

The website provides:

Stable Release (recommended)
Feature Release (latest development version)

Example release:

LAMMPS Stable Release – 22 Jul 2025

You can also download using Git:

git clone https://github.com/lammps/lammps.git
cd lammps

6. Create Build Directory¶

LAMMPS uses an out-of-source build to keep compiled files separate from the source code.

mkdir build
cd build

7. Configure LAMMPS with GPU Support¶

Run the following CMake command to enable GPU acceleration:

cmake ../cmake \
-D BUILD_MPI=yes \
-D PKG_GPU=yes \
-D GPU_API=cuda \
-D GPU_ARCH=sm_80

Explanation of Flags¶

Flag	Meaning
`BUILD_MPI=yes`	Enables MPI parallel computing
`PKG_GPU=yes`	Enables the LAMMPS GPU package
`GPU_API=cuda`	Uses NVIDIA CUDA for GPU acceleration
`GPU_ARCH=sm_80`	Optimizes the build for the specific GPU architecture

GPU Architecture Examples¶

GPU	Architecture Flag
Tesla V100	`sm_70`
Tesla A100	`sm_80`
RTX 3090	`sm_86`

8. Compile LAMMPS¶

Compile the software using multiple CPU cores:

make -j 48

Explanation:

-j 48 allows compilation using 48 CPU cores, which speeds up the build process.

9. Verify GPU Support¶

After compilation, verify that the GPU package is enabled:

./lmp -h | grep GPU

Expected output:

GPU package installed

10. Run a GPU Test Simulation¶

Run a simple GPU simulation test:

mpirun -np 16 ./lmp -sf gpu -pk gpu 1 -in in.lj

Explanation:

Option	Meaning
`-sf gpu`	Enables GPU-accelerated styles
`-pk gpu 1`	Uses one GPU device
`-np 16`	Runs the simulation on 16 CPU cores

11. Monitor GPU Usage¶

While the simulation is running, verify GPU utilization:

nvidia-smi

You should see LAMMPS processes using GPU memory and compute resources.

Advantages of GPU-Accelerated LAMMPS¶

GPU acceleration offers several benefits:

10×–100× faster simulations
Faster pair interaction calculations
Improved performance for large-scale molecular dynamics simulations

This is particularly useful for:

Large atomistic systems
Long simulation times
Materials science and nanotechnology research

Example HPC GPU Execution¶

Example production run on a GPU node:

mpirun -np 48 lmp -sf gpu -pk gpu 2 -in input.in

Resource usage:

Resource	Value
CPU Cores	48
GPUs	2
MPI Processes	48

Simple GPU Test Input Script¶

Example LAMMPS input file used for GPU testing:

units lj
atom_style atomic
lattice fcc 0.8442
region box block 0 20 0 20 0 20
create_box 1 box
create_atoms 1 box

pair_style lj/cut 2.5
pair_coeff 1 1 1.0 1.0 2.5

velocity all create 1.0 12345
fix 1 all nve

run 5000

Important Notes¶

Before compiling LAMMPS with GPU support, verify the following:

GPU driver is installed correctly
CUDA version is compatible with the GPU driver
MPI libraries are properly installed

Incorrect configuration of these components may cause GPU build or runtime failures.