GNNerator: A Hardware/Software Framework for Accelerating Graph Neural Networks

TL;DR

GNNerator is a specialized hardware/software framework that accelerates graph neural network computations by optimizing for dense and sparse operations, achieving significant speedups over existing solutions.

Contribution

It introduces GNNerator, a heterogeneous accelerator with a novel feature-blocking dataflow tailored for GNNs, improving performance and efficiency.

Findings

Achieves 5.7-37x speedup over NVIDIA RTX 2080-Ti.

Achieves 2.3-3.8x speedup over HyGCN.

Effectively balances irregular and regular memory accesses.

Abstract

Graph Neural Networks (GNNs) use a fully-connected layer to extract features from the nodes of a graph and aggregate these features using message passing between nodes, combining two distinct computational patterns: dense, regular computations and sparse, irregular computations. To address this challenge, we propose GNNerator, an accelerator with heterogeneous compute engines optimized for these two patterns. Further, GNNerator implements feature-blocking, a novel GNN dataflow that beneficially trades off irregular memory accesses during aggregation for regular memory accesses during feature extraction. We show GNNerator achieves speedups of 5.7-37x over an NVIDIA RTX 2080-Ti, and 2.3x-3.8x over HyGCN, a state-of-the-art GNN accelerator.