Update installation.rst

docs/installation.rst

@@ -10,28 +10,66 @@ Clone the repository
         git clone https://github.com/FarrOutLab/GWInferno.git
         cd gwinferno
 
-Recommended to use conda to set up your environment with python versions newer than at least 3.9:
+It is recommended to use conda to set up your environment with python versions newer than at least 3.12:
 
-For CPU usage only create an environment and install requirements and GWInferno with:
+---------------------------------------------------------------------
+For CPU
+---------------------------------------------------------------------
+
+For CPU usage only, create an environment and install requirements and GWInferno with:
 
 .. code-block:: bash
 
         cd gwinferno
-        conda create -n gwinferno python=3.10
+        conda create -n gwinferno python=3.12
         conda activate gwinferno
         conda install -c conda-forge numpyro h5py 
         pip install --upgrade pip
         pip install -r pip_requirements.txt
         python -m pip install .
 
-To enable JAX access to CUDA enabled GPUs we need to specify specific versions to install (See `JAX <https://github.com/google/jax>`_ installation instructions for more details). For a GPU enabled environment use:
+---------------------------------------------------------------------
+For GPU
+---------------------------------------------------------------------
+
+To enable JAX access to CUDA enabled GPUs, we need to specify specific versions to install. The following procedure will only work for Linux x86_64 and Linux aarch64; for other platforms see `Jax documentation <https://jax.readthedocs.io/en/latest/installation.html>`_.
+
+Jax recommends installing Nvidia CUDA and cuDNN with pip wheels. If you use local installations of CUDA and cuDNN, which could be the case for a remote cluster, then you'll need to install jax from the single CUDA wheel variant it offers. As of writing, this wheel is only compatible with CUDA >= 12.1 and cuDNN >= 9.1 < 10.0. See `JAX <https://github.com/google/jax>`_ installation instructions for more details.
+
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+Installation process for CUDA installed via pip:
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+
+.. code-block:: bash
+
+        cd gwinferno
+        conda create -n gwinferno_gpu python=3.12
+        conda activate gwinferno_gpu
+        pip install --upgrade pip
+
+Next, install CUDA and cuDNN pip wheels. See `Nvidia's CUDA quickstart guide <https://docs.nvidia.com/cuda/cuda-quick-start-guide/index.html#pip-wheels-linux>`_ for the CUDA installation procedure and the `cuDNN documentation <https://docs.nvidia.com/deeplearning/cudnn/latest/installation/linux.html#installing-cudnn-with-pip>`_ for the cuDNN installation procedure. Once that has finished, continue with these steps:
+
+.. code-block:: bash
+
+        pip install --upgrade "jax[cuda12]"
+        pip install numpyro[cuda]
+        pip install -r pip_requirements.txt
+        python -m pip install .
+
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+Installation process for locally installed CUDA and cuDNN:
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Ensure `Nvidia CUDA <https://docs.nvidia.com/cuda/cuda-quick-start-guide/index.html#linux>`_ and `cuDNN <https://docs.nvidia.com/deeplearning/cudnn/latest/installation/linux.html>`_ are installed locally.
 
 .. code-block:: bash
 
         cd gwinferno
-        conda create -n gwinferno_gpu python=3.10
+        conda create -n gwinferno_gpu python=3.12
         conda activate gwinferno_gpu
-        conda install -c nvidia -c conda-forge jaxlib=*=*cuda* jax cuda-nvcc numpyro h5py
         pip install --upgrade pip
+        pip install --upgrade "jax[cuda12_local]"
+        pip install numpyro[cuda]
         pip install -r pip_requirements.txt
         python -m pip install .