Graph Neural Network Frameworks
Graph neural network (GNN) frameworks are easy-to-use Python packages that offer building blocks to build GNNs on top of existing deep learning frameworks for a wide range of applications.
NVIDIA AI Accelerated GNN frameworks are optimized to deliver high-performance preprocessing, sampling, and training on NVIDIA GPUs.
Explore the benefits.
NVIDIA AI optimized GNN frameworks.
GNN framework containers for Deep Graph Library (DGL) and PyTorch Geometric (PyG) come with the latest NVIDIA RAPIDs, PyTorch, and frameworks that are performance tuned and tested for NVIDIA GPUs.
Go from hours to minutes. With NVIDIA RAPIDS™ integration, cuDF accelerates pandas queries up to 39X faster than CPU so that you can run ETL with GPU-optimized code.
End-to-end reference examples.
Streamline workflows for GNNs, from experimentation to production, with GPU-optimized, tested, and validated examples for fraud detection, recommender systems, and drug discovery.
Runs on prem and in cloud.
Preprocess, train, and run inference on graph neural networks on multi-GPU, multi-node infrastructure.
See performance benchmarks.
With NVIDIA AI Accelerated GNN frameworks, you can get end-to-end performance optimization, making it the fastest solution to preprocess and build GNNs.
DGL Container, Dataset: MAG240M, Model: RCGN, Total edges: 1.7B
GPU: 1x A100 80GB, CPU: AMD EPYC 7742 64-Core
NVIDIA AI Accelerated GNN frameworks.
Deep Graph Library
Deep Graph Library (DGL) is an easy-to-use and scalable Python library used for implementing and training GNNs.
To enable developers to quickly take advantage of GNNs, we’ve partnered with the DGL team to provide a containerized solution that includes the latest DGL, PyTorch, and NVIDIA RAPIDS (cuDF, XGBoost, RMM, cuML, and cuGraph), which can be used to accelerate ETL operations and training.
Our private early access release includes two containers:
- A ready-to-use DGL container with the latest upstream improvements and tested dependencies.
- A ready-to-use DGL container with tested dependencies, an optimized SE(3)-Transformer model, and an accelerated neural network training environment based on DGL and PyTorch. The SE(3)-Transformer for DGL container is suited for recognizing three-dimensional shapes making it useful for segmenting lidar point clouds or in pharmaceutical and drug discovery research.
Apply for early access to our DGL container or the SE(3)-Transformer for DGL container.
Apply for Early Access
PyTorch Geometric (PyG) is a library built upon PyTorch to easily write and train GNNs for a wide range of applications related to structured data.
We've collaborated with the PyG team to offer the best performance on NVIDIA GPUs—a containerized solution that includes the latest version of PyG, PyTorch, and NVIDIA RAPIDS, which can be used to speed up the GNN workflow on GPUs.
A ready-to-use PyG container with the latest upstream improvements and tested dependencies will be available in Q4’2022 in private early access.
Apply for Early Access
“OrbNet, with the help of DGL and NVIDIA GPUS, has enabled the accurate and data-efficient prediction of drug molecule properties, reducing by years the amount of time needed to advance new drug candidates through lead identification and lead optimization.”
— Tom Miller PhD, Co-founder and CEO, Entos
“Pinterest uses graph neural networks with billions of nodes and edges to understand our ecosystem of over 300 billion pins. We rely on GPUs and NVIDIA-optimized libraries for the training and inference of these models.”
— Andrew Zhai, Senior Machine Learning Architect, Pinterest
“AMEX research is excited for DGL to help improve their cardholder experiences
through improved fraud detection.”
— Aziza Johnson, VP Corporate Affairs and Communication, American Express
"NVIDIA’s end-to-end GNN solutions perfectly integrate PyTorch and DGL, allowing us to analyze the datasets of one of the world's largest material models (290M atoms). Our data process pipeline is benefited greatly from NVIDIA RAPIDs, and significantly reduced by 80%. NVIDIA DGL Container also enables triple faster GNN model training and doubles the inference efficiency."
— Nien-Ti Tsou, Associate Professor, Department of Materials Science and Engineering, National Chiao Tung University
"Meituan's GNN platform, with optimizations to DGL and GPU performance, serves many services at Meituan including search, recommendation, advertising, etc."
— Mengdi Zhang, Head of Graph Learning, Senior Algorithm Expert, Meituan
How GNNs are being used across industries
Fraud detection in financial services.
Fraudulent transactions are some of the most serious threats to the financial services industry. GNNs have emerged as a powerful tool to improve fraud detection tasks where fraudulent transactions are identified by aggregating neighbor information.
Drug discovery in healthcare.
To develop new candidate medications quickly, researchers need to compute the physical properties of molecules. GNNs provide an efficient way to represent these complex 3D structures and accurately predict their properties.
Recommenders in retail.
GNNs can leverage information from existing data with a graph structure, resulting in better predictions than traditional methods. Recommendation systems use a form of node embeddings in GNNs to match customers with products.
Get started with DGL.
Check out tutorials on how to get started with the NVIDIA DGL Container and walk through examples of preprocessing, training, and inference.