If you want to utilize the full set of features from :pyg:`PyG`, there exists several additional libraries you may want to install:

* `pyg-lib <https://github.com/pyg-team/pyg-lib>`__: Heterogeneous GNN operators, graph sampling routines, and :class:`~torch_geometric.nn.conv.SplineConv` support
* `torch-scatter <https://github.com/rusty1s/pytorch_scatter>`__: Accelerated and efficient sparse reductions
* `torch-sparse <https://github.com/rusty1s/pytorch_sparse>`__: :class:`SparseTensor` support, see `here <https://pytorch-geometric.readthedocs.io/en/latest/advanced/sparse_tensor.html>`__
* `torch-cluster <https://github.com/rusty1s/pytorch_cluster>`__: Graph clustering routines

.. note::
   ``torch-spline-conv`` is no longer required as a separate package.
   Its functionality has been migrated to ``pyg-lib>=0.6.0``.

These packages come with their own CPU and GPU kernel implementations based on the :pytorch:`null` `PyTorch C++/CUDA/hip(ROCm) extension interface <https://github.com/pytorch/extension-cpp/>`_.
For a basic usage of :pyg:`PyG`, these dependencies are **fully optional**.
We recommend to start with a minimal installation, and install additional dependencies once you start to actually need them.

Installation from Wheels

  python -c "import torch; print(torch.__version__)"
  >>> 2.10.0

  python -c "import torch; print(torch.version.cuda)"
  >>> 12.8

  pip install pyg_lib torch_scatter torch_sparse torch_cluster -f https://data.pyg.org/whl/torch-${TORCH}+${CUDA}.html

  pip install pyg_lib torch_scatter torch_sparse torch_cluster -f https://data.pyg.org/whl/torch-2.10.0+cu130.html

  pip install pyg_lib torch_scatter torch_sparse torch_cluster -f https://data.pyg.org/whl/torch-2.9.0+cu128.html

In case a specific version is not supported by `our wheels <https://data.pyg.org/whl>`_, you can alternatively install them from source:

#. Ensure that your CUDA is setup correctly (optional):

   #. Check if :pytorch:`PyTorch` is installed with CUDA support:

      .. code-block:: none

         python -c "import torch; print(torch.cuda.is_available())"
         >>> True

   #. Add CUDA to :obj:`$PATH` and :obj:`$CPATH` (note that your actual CUDA path may vary from :obj:`/usr/local/cuda`):

      .. code-block:: none

         export PATH=/usr/local/cuda/bin:$PATH
         echo $PATH
         >>> /usr/local/cuda/bin:...

         export CPATH=/usr/local/cuda/include:$CPATH
         echo $CPATH
         >>> /usr/local/cuda/include:...

   #. Add CUDA to :obj:`$LD_LIBRARY_PATH` on Linux and to :obj:`$DYLD_LIBRARY_PATH` on macOS (note that your actual CUDA path may vary from :obj:`/usr/local/cuda`):

      .. code-block:: none

         export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
         echo $LD_LIBRARY_PATH
         >>> /usr/local/cuda/lib64:...

         export DYLD_LIBRARY_PATH=/usr/local/cuda/lib:$DYLD_LIBRARY_PATH
         echo $DYLD_LIBRARY_PATH
         >>> /usr/local/cuda/lib:...

   #. Verify that :obj:`nvcc` is accessible from terminal:

      .. code-block:: none

         nvcc --version
         >>> 11.8

   #. Ensure that :pytorch:`PyTorch` and system CUDA versions match:

      .. code-block:: none

         python -c "import torch; print(torch.version.cuda)"
         >>> 11.8

         nvcc --version
         >>> 11.8

#. Install the relevant packages:

   .. code-block:: none

      pip install --verbose git+https://github.com/pyg-team/pyg-lib.git
      pip install --verbose torch_scatter
      pip install --verbose torch_sparse
      pip install --verbose torch_cluster

In rare cases, CUDA or :python:`Python` path problems can prevent a successful installation.
:obj:`pip` may even signal a successful installation, but execution simply crashes with :obj:`Segmentation fault (core dumped)`.
We collected common installation errors in the `Frequently Asked Questions <installation.html#frequently-asked-questions>`__ subsection.
In case the FAQ does not help you in solving your problem, please create an `issue <https://github.com/pyg-team/pytorch_geometric/issues>`_.
Before, please verify that your CUDA is set up correctly by following the official `installation guide <https://docs.nvidia.com/cuda>`_.

Installation via Anaconda
-------------------------

.. warning::
   Conda packages are no longer available since :pytorch:`PyTorch` :obj:`>2.5.0`.
   Please use :obj:`pip` instead.

For earlier :pytorch:`PyTorch` versions (:obj:`torch<=2.5.0`), you can install :pyg:`PyG` via :conda:`null` `Anaconda <https://anaconda.org/pyg/pyg>`_ for all major OS, and CUDA combinations.
If you have not yet installed :pytorch:`PyTorch`, install it via :conda:`null` :obj:`conda install` as described in its `official documentation <https://pytorch.org/get-started/locally/>`_.
Given that you have :pytorch:`PyTorch` installed, run

.. code-block:: none

   conda install pyg -c pyg

If :conda:`null` :obj:`conda` does not pick up the correct CUDA version of :pyg:`PyG`, you can enforce it as follows:

.. code-block:: none

   conda install pyg=*=*cu* -c pyg

.. _install-cugraph:

Accelerating PyG with NVIDIA cuGraph GNN
----------------------------------------

:pyg:`PyG` can optionally leverage NVIDIA's `cuGraph <https://github.com/rapidsai/cugraph>`_ to accelerate neighbor sampling and achieve better scalability for multi-GPU training on large-scale graphs (2x-8x data loading speedups on billion-edge graphs).

NVIDIA currently recommends the `NVIDIA PyG Container <https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pyg>`_ on NGC as the most reliable way to use cuGraph integration with :pyg:`PyG`. For other installation methods, refer to the `cuGraph GNN repository <https://github.com/rapidsai/cugraph-gnn>`_ and/or the `RAPIDS installation guide <https://docs.rapids.ai/install>`_.

.. note::

   **cuGraph GNN is optional** — all :pyg:`PyG` functionality, including multi-GPU training, works without it. However, for users with NVIDIA GPUs, cuGraph can provide significant speedups and better scalability for neighbor sampling and data loading, especially on large-scale graphs.

`cuGraph <https://github.com/rapidsai/cugraph>`_ is a collection of packages focused on GPU-accelerated graph analytics including support for property graphs and scaling up to thousands of GPUs. cuGraph supports the creation and manipulation of graphs followed by the execution of scalable fast graph algorithms. It is part of the `RAPIDS <https://rapids.ai>`_ accelerated data science framework.

`cuGraph GNN <https://github.com/rapidsai/cugraph-gnn>`_ is a collection of GPU-accelerated plugins that support :pytorch:`PyTorch` and :pyg:`PyG` natively through the *cuGraph-PyG* and *WholeGraph* subprojects. cuGraph GNN is built on top of cuGraph, leveraging its low-level `pylibcugraph <https://github.com/rapidsai/cugraph/tree/branch-25.08/python/pylibcugraph>`_ API and C++ primitives for sampling and other GNN operations (`libcugraph <https://github.com/rapidsai/cugraph/tree/branch-25.08/python/libcugraph>`_). It also includes the :obj:`libwholegraph` and :obj:`pylibwholegraph` libraries for high-performance distributed edgelist and embedding storage. Users have the option of working with these lower-level libraries directly, or through the higher-level API in cuGraph-PyG that directly implements the :class:`~torch_geometric.data.GraphStore`, :class:`~torch_geometric.data.FeatureStore`, :class:`~torch_geometric.loader.NodeLoader`, and :class:`~torch_geometric.loader.LinkLoader` interfaces.

Complete documentation on RAPIDS graph packages, including ``cugraph``, ``cugraph-pyg``, ``pylibwholegraph``, and ``pylibcugraph`` is available on the `RAPIDS docs pages <https://docs.rapids.ai/api/cugraph/nightly/graph_support>`_.

See `rapidsai/cugraph-gnn examples on GitHub <https://github.com/rapidsai/cugraph-gnn/tree/branch-25.12/python/cugraph-pyg/cugraph_pyg/examples>`_ for fully scalable PyG example workflows.

Frequently Asked Questions
--------------------------

#. :obj:`undefined symbol: **make_function_schema**`: This issue signals (1) a **version conflict** between your installed :pytorch:`PyTorch` version and the :obj:`${TORCH}` version specified to install the extension packages, or (2) a version conflict between the installed CUDA version of :pytorch:`PyTorch` and the :obj:`${CUDA}` version specified to install the extension packages.
   Please verify that your :pytorch:`PyTorch` version and its CUDA version **match** with your installation command:

   .. code-block:: none

      python -c "import torch; print(torch.__version__)"
      python -c "import torch; print(torch.version.cuda)"
      nvcc --version

   For re-installation, ensure that you do not run into any caching issues by using the :obj:`pip --force-reinstall --no-cache-dir` flags.
   In addition, the :obj:`pip --verbose` option may help to track down any issues during installation.
   If you still do not find any success in installation, please try to install the extension packages `from source <installation.html#installation-from-source>`__.