GPU4PyScf - BNNLab/BN_Group_Wiki GitHub Wiki
On your own computer
It is strongly advisable to create a conda environment for this.
conda create --name gpu4pyscf-env python=3.11
conda activate gpu4pyscf-env
The package at the moment is not compatible with the latest version of Python, so 3.11 version is recommended.
Follow the first section of instructions on the GPU4pyscf Github:
[!NOTE] The compiled binary packages support compute capability 6.0 and later (Pascal and later, such as Tesla P100, RTX 10 series and later).
Run nvcc --version in your terminal to check the installed CUDA toolkit version. Then, choose the proper package based on your CUDA toolkit version.
| Platform | Command | cutensor (highly recommended) |
|---|---|---|
| CUDA 11.x | pip3 install gpu4pyscf-cuda11x |
pip3 install cutensor-cu11 |
| CUDA 12.x | pip3 install gpu4pyscf-cuda12x |
pip3 install cutensor-cu12 |
On AIRE
There is one key change on AIRE, due to CUDA toolkit being available as a module. You will need to load this module first before proceeding with the instructions above.