DGL Release 23.11

The NVIDIA container image for DGL, release 23.11, is available on NGC.

Contents of the DGL container

This container image contains the complete source of the version of DGL in /opt/dgl/dgl-source. It is pre-built and installed as a system Pyton module. The container includes the following:

• DGL 1.2+4b6faec
• PyTorch 23.11
• RAPIDS 23.10
• NVIDIA CUDA® 12.3.0
• NVIDIA cuBLAS 12.2.5.6
• NVIDIA cuDNN 8.9.6
• NVIDIA NCCL 2.19.3
• Apex
• rdma-core 39.0
• NVIDIA HPC-X 2.16
• OpenMPI 4.1.4+
• GDRCopy 2.3
• TensorBoard 2.12.0
• Nsight Compute 2023.3.0.12
• Nsight Systems 2023.3.1.92
• NVIDIA TensorRT™ 8.6.1.6
• Torch-TensorRT 1.4.0
• NVIDIA DALI® 1.31.0
• MAGMA 2.6.2
• JupyterLab 2.3.2 including Jupyter-TensorBoard
• PyTorch quantization wheel v2.1.2
• TransformerEngine 0.3.0
• NVSHMEM 2.10.1

GPU Requirements

Release 23.11 supports CUDA compute capability 6.0 and later. This corresponds to GPUs in the NVIDIA Pascal, NVIDIA Volta™, NVIDIA Turing™, NVIDIA Ampere architecture, and NVIDIA Hopper™ architecture families. For a list of GPUs to which this compute capability corresponds, see CUDA GPUs. For additional support details, see Deep Learning Frameworks Support Matrix.

Key Features and Enhancements

This DGL release includes the following key features and enhancements.

• DGL container image version 23.11 is based on DGL 1.1.1.
• In this release of the NVIDIA DGL container, we enhance support for distributed feature gathering by integrating NVSHMEM, further improving on the feature fetching performance for distributed GNN tasks. Check out the examples located at: /workspace/examples/wholegraph-examples
• Add NVIDIA Synthetic Graph Generation tool for generating graphs with an arbitrary size, including node and edge tabular features.

  The major features of the release can be found in the DGL release notes.

Announcements

None.

NVIDIA DGL Container Versions

The following table shows what versions of Ubuntu, CUDA, DGL, and TensorRT are supported in each NVIDIA containers for DGL. For older container versions, refer to the Frameworks Support Matrix.

Known Issues

• When cpu sampling is enabled (use_uva=False and num_workers>0), DGL sampling process would initialize cuda instance (issue-6561), which could result in a segmentation fault with the current cuda driver in the container.

• The tensors that are used as node features must be contiguous and cannot be views of other tensors when the use_uva flag is set to True in the dgl.dataloading.Dataloader class.

  When you attempt to use a graph with a non-contiguous or view tensors for edata or ndata, aDGLError will occur.