Installing the GPU Module

The GPU ray tracing module is still an experimental feature and is under active development. If you encounter any issues or have suggestions, please let us know on GitHub.

Requirements

The GPU ray tracing module is implemented using OptiX 8.0. To use it, ensure your system meets the following requirements:

  • NVIDIA Driver: Version 535 or higher
  • CUDA Toolkit: Version 12.0
  • OptiX SDK: Version 8.0

CMake Configuration

To enable GPU support, follow these steps:

  1. Run CMake with the -DVIENNAPS_USE_GPU=ON option to enable GPU support.
  2. Specify the path to your OptiX installation by setting the CMake variable OptiX_INSTALL_DIR.
    • This should point to the directory where the include folder containing OptiX headers is located. E.g.: 
      cmake -DVIENNAPS_USE_GPU=ON -DOptiX_INSTALL_DIR=/path/to/optix .

      Alternatively, you can set the OptiX_INSTALL_DIR environment variable:

      export OptiX_INSTALL_DIR=/path/to/optix

      This will be used during the CMake setup.

  3. Install the CUDA toolkit system-wide so CMake can detect it automatically or provide CUDA_PATH CMake variable.
  4. Run CMake configuration. If both CUDA and OptiX are found, the GPU extension will be enabled.
  5. (Optional) To build examples or tests, set:
    • VIENNAPS_BUILD_EXAMPLES=ON
    • VIENNAPS_BUILD_TESTS=ON

Python Build Instructions

For building the Python GPU module:

  1. Set the environment variable OptiX_INSTALL_DIR to the OptiX installation directory.
    • This variable will be used during the CMake setup.
  2. Run the following command to install the module locally: 
    CMAKE_ARGS=-DVIENNAPS_USE_GPU=ON pip install .
  3. The GPU functions are available in the GPU submodule: 
    import viennaps3d as vps
context = vps.gpu.Context()
context.create()
gpuProcess = vps.gpu.Process(context)

    Note: The GPU submodule is only available in the 3D bindings since GPU ray tracing is implemented for 3D only.

Example usage of the GPU module can be found in the examples.

Currently, only a limited number of models are available for GPU acceleration: SingleParticleProcess, MultiParticleProcess, and SF6O2Etching.