Setup CONDA Environments with Full CUDA Support

Setting up a conda environment with cuda support seems to be somewhat straight forward. The getting started guide of PyTorch is straight forward regarding this. A new environment is created or an existing one is activated and PyTorch is installed as follows:

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

So far, so good. While we may get along with it, it certainly is not full cuda and cudnn support. Some python packages may require PyCUDA or CuPy (or other) or we want to use one of these packages ourselves. Depending what custom kernels are used/deployed, libcublas-dev or other libraries are required. In order to install them, the full CUDA package would be installed with:

conda install -c nvidia cuda-libraries-dev cudatoolkit