Hello Everyone
I am trying to install LAMMPS from Conda-Forge to perform simulations using M3GNet. However, as you may know, the version of LAMMPS available on Conda-Forge doesn’t support parallelization via OpenMPI. Therefore, I would like to utilize the GPU for my simulations. I have already installed the necessary GPU drivers on my system.
Here is the information for the CUDA binary:
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:02:13_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0
The steps I followed for installing LAMMPS from Conda-Forge, as well as M3GNet, MatGl, DGL, and PyTorch, are as follows:
Create and activate a new Conda environment:
conda create -n lammps_m3gnet python=3.11
conda activate lammps_m3gnet
Install LAMMPS from Conda-Forge:
conda install -c conda-forge lammps
Install M3GNet and MatGl:
pip install m3gnet
pip install matgl
Install CUDA Toolkit (v11.8.0):
conda install -c conda-forge cudatoolkit=11.8.0
Install PyTorch compatible with matgl:
pip install torch==2.2.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
Install DGL:
pip install dgl -f https://data.dgl.ai/wheels/cu118/repo.html
pip install dglgo -f https://data.dgl.ai/wheels-test/repo.html
After completing the installation, when I run a job using the LAMMPS binary (lmp), I get the following output:
PID USER DEV TYPE GPU GPU MEM CPU HOST MEM Command
3923427 ss 0 Compute 0% 386MiB 2% 100% 6742MiB lmp -in input.txt
3928976 ss 0 Compute 0% 386MiB 2% 100% 4725MiB lmp -in input.txt
As you can see, only 2% of the GPU memory is being utilized for each job. My question is: How can I increase the percentage of GPU memory used by each job?
I appreciate your assistance in advance.