EnGN: A High-Throughput and Energy-Efficient Accelerator for Large Graph Neural Networks

Liang, Shengwen; Wang, Ying; Liu, Cheng; He, Lei; Li, Huawei; Xu, Daolin; Li, Xiaowei

doi:10.1109/tc.2020.3014632

Cited by 104 publications

(90 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The main challenge in GNN accelerator design is the alternation of phases with either dense or extremely sparse computation. The sparsity is driven by the graph connectivity or the graph adjacency matrix [16]- [18]. On the other hand, phases of dense computations are usually due to the dense nature of the operations that are applied to the nodes and edges in parallel [16]- [18].…”

Section: Acceleratorsmentioning

confidence: 99%

“…The sparsity is driven by the graph connectivity or the graph adjacency matrix [16]- [18]. On the other hand, phases of dense computations are usually due to the dense nature of the operations that are applied to the nodes and edges in parallel [16]- [18]. Additionally, GNNs process input graphs that might have billions of nodes and edges, with uneven connectivity.…”

Section: Acceleratorsmentioning

confidence: 99%

“…Given their distinct architectures, in this work we focus on two state-of-the-art GNN acceleration engines, i.e., EnGN [16] and HyGCN [17]. Figs.…”

Section: Acceleratorsmentioning

confidence: 99%

See 2 more Smart Citations

Characterizing the Communication Requirements of GNN Accelerators: A Model-Based Approach

Guirado

Jain

Abadal

et al. 2021

2021 IEEE International Symposium on Circuits and Systems (ISCAS)

View full text Add to dashboard Cite

Relational data present in real world graph representations demands for tools capable to study it accurately. In this regard Graph Neural Network (GNN) is a powerful tool, wherein various models for it have also been developed over the past decade. Recently, there has been a significant push towards creating accelerators that speed up the inference and training process of GNNs. These accelerators, however, do not delve into the impact of their dataflows on the overall data movement and, hence, on the communication requirements. In this paper, we formulate analytical models that capture the amount of data movement in the most recent GNN accelerator frameworks. Specifically, the proposed models capture the dataflows and hardware setup of these accelerator designs and expose their scalability characteristics for a set of hardware, GNN model and input graph parameters. Additionally, the proposed approach provides means for the comparative analysis of the vastly different GNN accelerators.

show abstract

Section: Acceleratorsmentioning

confidence: 99%

Section: Acceleratorsmentioning

confidence: 99%

See 1 more Smart Citation

Characterizing the Communication Requirements of GNN Accelerators: A Model-Based Approach

Guirado

Jain

Abadal

et al. 2021

2021 IEEE International Symposium on Circuits and Systems (ISCAS)

View full text Add to dashboard Cite

show abstract

“…In order to support more GNN models, Liang, etc. [7] proposed a fine-grained GNN processing model and developed a corresponding GNN accelerator EnGN. They had a 2D computing array combined with Automatic accelerator generation.…”

Section: Related Workmentioning

confidence: 99%

“…Although specialized GNN accelerators of ASICs such as HyGCN [26] and EnGN [7], are tailored as alternatives to CPU and GPU, they are still not flexible enough to meet the requirements of various GNN-based applications from the cloud to edge. First and foremost, because graph learning is a fast-developing field and novel GNN architectures keep emerging, ASIC-based accelerators of fixed architecture sometimes cannot run state-of-the-art GNN models.…”

Section: Introductionmentioning

confidence: 99%