A General and Scalable GCN Training Framework on CPU Supercomputers
Graph Convolutional Networks (GCNs) are widely used in various domains. However, training distributed full-batch GCNs on large-scale graphs poses challenges due to inefficient memory access patterns and high communication overhead. This paper presents a general and efficient GCN training framework on CPU supercomputers. It comprises a general aggregation kernel designed to optimize irregular memory access and a quantization method with label propagation to reduce communication overhead. Experimental results show that our method achieves a speedup of up to 4.1× compared with the SoTA implementations.
Authors
- Mohamed Wahib (ORCID: https://orcid.org/0000-0002-7165-2095)
- Toshio Endo (ORCID: https://orcid.org/0000-0001-7297-6211)
- Xin Liu (ORCID: https://orcid.org/0000-0002-2336-7409)
- Rio Yokota (ORCID: https://orcid.org/0000-0001-7573-7873)
- Zhuang Chen (ORCID: https://orcid.org/0009-0006-4156-9879)
- Nikoli Dryden
- Lingqi Zhang (ORCID: https://orcid.org/0000-0002-2452-1551)
- Peng Chen (ORCID: https://orcid.org/0000-0003-1244-3151)
- Satoshi Matsuoka (ORCID: https://orcid.org/0000-0003-1910-8532)
Institutions
- Lawrence Livermore National Laboratory (US)
- Shanghai Institute for Science of Science (CN)
- RIKEN Center for Computational Science (JP)
- Institute of Science Tokyo (JP)
- National Institute of Advanced Industrial Science and Technology (JP)
Publication Details
- Published
- 2025-02-28
- DOI
- https://doi.org/10.1145/3710848.3710860
- Primary Topic
- Brain Tumor Detection and Classification
- Type
- article
- Field-Weighted Citation Impact
- 0.00