MaxK-GNN: Towards Theoretical Speed Limits for Accelerating Graph Neural Networks Training

arXiv:2312.08656v3 Announce Type: replace
Abstract: In the acceleration of deep neural network training, the GPU has become the mainstream platform. GPUs face substantial challenges on GNNs, such as workload imbalance and memory access irregularities, leading to underutilized hardware. Existing solutions such as PyG, DGL with cuSPARSE, and GNNAdvisor frameworks partially address these challenges but memory traffic is still significant.
We argue that drastic performance improvements can only be achieved by the vertical optimization of algorithm and system innovations, rather than …

abstract arxiv become challenges cs.ai cs.dc cs.lg deep neural network dgl face frameworks gnn gnns gpu gpus graph graph neural networks hardware memory network networks network training neural network neural networks platform solutions speed training type