BitGNN: Unlocking the Performance Potential of Binary Graph Neural Networks on GPUs

Conference · Fri Jun 23 04:00:00 EDT 2023

DOI:https://doi.org/10.1145/3577193.3593725· OSTI ID:2203530

Chen, Jou-An ^[1]; Sung, Hsin-Hsuan ^[1]; Shen, Xipeng ^[1]; Choudhury, Sutanay ^[2]; Li, Ang ^[2]

---

1. North Carolina State University
2. BATTELLE (PACIFIC NW LAB)

Graph Neural Networks (GNNs) have shown compelling results in many graph-based learning tasks. They are, however, time-consuming. Recent work has shown a promising direction in improving GNN speed and shrinking the size — network binarization, which binarizes network values and operations. Prior work, however, mainly focused on algorithm designs, leaving it open on how to fully materialize the performance potential. This work fills the gap by proposing techniques to best map binary GNNs and their computations to fit the nature of bit manipulations, optimizations and algorithms to maximize BSpMM kernel efficiency, and solutions to other factors influencing the end-to-end time on GPUs. Results on real-world graphs show that the proposed techniques outperform state of-the-art binary GNN implementations by 21-67× with little accuracy loss.

• Research Organization: Pacific Northwest National Laboratory (PNNL), Richland, WA (United States)
• Sponsoring Organization: USDOE
• DOE Contract Number: AC05-76RL01830
• OSTI ID: 2203530
• Report Number(s): PNNL-SA-171811
• Country of Publication: United States
• Language: English