SuperGCN trains full-batch GCNs on CPU supercomputers with minimum-vertex-cover hybrid aggregation and Int2 quantization, reaching 8,192 MPI ranks and beating prior CPU and GPU frameworks on several large datasets.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.DC 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Scaling Large-scale GNN Training to Thousands of Processors on CPU-based Supercomputers
SuperGCN trains full-batch GCNs on CPU supercomputers with minimum-vertex-cover hybrid aggregation and Int2 quantization, reaching 8,192 MPI ranks and beating prior CPU and GPU frameworks on several large datasets.