SNI-GNN system uses SmartNICs to accelerate full-graph GNN training

Researchers have developed SNI-GNN, a system designed to improve the efficiency of full-graph Graph Neural Network (GNN) training on large clusters. This system utilizes SmartNICs to predict remote embeddings in-network, thereby reducing communication overhead. SNI-GNN incorporates a lightweight predictor on SmartNICs, an importance-based sampling policy, and an asynchronous data pipeline, demonstrating significant speedups and communication reductions with minimal accuracy loss. AI

IMPACT This system could enable more efficient training of large-scale GNNs, potentially accelerating research and deployment in areas like recommendation systems and drug discovery.

RANK_REASON The cluster contains a research paper detailing a novel system for GNN training. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• BNS-GCN
• Graph Neural Network
• NVIDIA BlueField-3
• SANCUS
• SmartNIC
• SNI-GNN

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →