Illinois Graph Benchmark: Large-Scale Real-World Data for GNNs

Illinois Graph Benchmark (IGB) is a collection of large-scale real-world graph datasets developed by the IBM-Illinois Discovery Accelerator Institute in collaboration with Amazon and NVIDIA. Updated in June 2025, it provides graph structures and node features specifically designed for benchmarking Graph Neural Network (GNN) performance. It is positioned as the largest open-source dataset of its kind for graph learning research.

Use Cases

• Node classification tasks using node features and adjacency matrices
• Link prediction based on graph edge connectivity patterns
• Benchmarking GNN training throughput and scalability on high-performance hardware

Strengths

• Developed through collaboration between IBM, Amazon, and NVIDIA Research
• Designed specifically for Graph Neural Network (GNN) benchmarking
• Open-source availability for large-scale graph learning research

Limitations

• High memory and compute requirements are likely given the scale of the data
• Specific feature schemas and label distributions are not detailed in the summary metadata

Provenance

Source

IBM-Illinois Discovery Accelerator Institute, Amazon Research, and NVIDIA Research

Collection Method

Aggregated real-world graph data

Freshness

Last updated June 2025.