GRIP: A Graph Neural Network Accelerator Architecture

Kiningham, Kevin; Ré, Christopher; Levis, Philip

doi:10.1109/tc.2022.3197083

Cited by 32 publications

(16 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 2 presents a baseline GCN accelerator comprising aggregation engines and combination engines, similar to previous work [8,1 Normalized Laplacian refers to a form of adjacency matrix defined as Ã = 𝐼 − 𝐷 −1/2 𝐴𝐷 −1/2 where 𝐷 is a degree matrix. 38,50,75], where the green component has been newly added that will be discussed in Section 4.3. The core of combination engine is a systolic array to support efficient matrix multiplication.…”

Section: Baseline Gcn Acceleratormentioning

confidence: 99%

Slice-and-Forge: Making Better Use of Caches for Graph Convolutional Network Accelerators

Yoo¹,

Song²,

Lee³

et al. 2023

Preprint

View full text Add to dashboard Cite

Graph convolutional networks (GCNs) are becoming increasingly popular as they can process a wide variety of data formats that prior deep neural networks cannot easily support. One key challenge in designing hardware accelerators for GCNs is the vast size and randomness in their data access patterns which greatly reduces the effectiveness of the limited on-chip cache. Aimed at improving the effectiveness of the cache by mitigating the irregular data accesses, prior studies often employ the vertex tiling techniques used in traditional graph processing applications. While being effective at enhancing the cache efficiency, those approaches are often sensitive to the tiling configurations where the optimal setting heavily depends on target input datasets. Furthermore, the existing solutions require manual tuning through trial-and-error or rely on sub-optimal analytical models.In this paper, we propose Slice-and-Forge (SnF), an efficient hardware accelerator for GCNs which greatly improves the effectiveness of the limited on-chip cache. SnF chooses a tiling strategy named feature slicing that splits the features into vertical slices and processes them in the outermost loop of the execution. This particular choice results in a repetition of the identical computational patterns over irregular graph data over multiple rounds. Taking advantage of such repetitions, SnF dynamically tunes its tile size. Our experimental results reveal that SnF can achieve 1.73× higher performance in geomean compared to prior work on multi-engine settings, and 1.46× higher performance in geomean on small scale settings, without the need for off-line analyses.

show abstract

Section: Baseline Gcn Acceleratormentioning

confidence: 99%

Slice-and-Forge: Making Better Use of Caches for Graph Convolutional Network Accelerators

Yoo¹,

Song²,

Lee³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…GNN systems adopt the idea of partitioning because the input graphs are unlikely to fit in a single machine's memory. Some systems use traditional edge-cut or vertex-cut methods [130,213] whereas others combine those with features like a cost model [87], feasibility score [111] or dataflow partitioning [95]. Table 3 summarizes the different partitioning methods.…”

Section: Partitioningmentioning

confidence: 99%

“…This partitioning strategy benefits the edge-wise processing, because only the source and destination vertex data need to be loaded. Unlike NeuGraph, GReTA [95] does not partition the graph itself, but the dataflow into blocks. The dataflow, also called nodeflow, is a graph structure representing the propagation of feature vectors throughout the forward pass of the GNN model.…”

Section: ✓ ✓ ✓mentioning

confidence: 99%

See 1 more Smart Citation

The Evolution of Distributed Systems for Graph Neural Networks and Their Origin in Graph Processing and Deep Learning: A Survey

Vatter,

Mayer,

Jacobsen

2023

ACM Comput. Surv.

View full text Add to dashboard Cite

Graph Neural Networks (GNNs) are an emerging research field. This specialized Deep Neural Network (DNN) architecture is capable of processing graph structured data and bridges the gap between graph processing and Deep Learning (DL). As graphs are everywhere, GNNs can be applied to various domains including recommendation systems, computer vision, natural language processing, biology and chemistry. With the rapid growing size of real world graphs, the need for efficient and scalable GNN training solutions has come. Consequently, many works proposing GNN systems have emerged throughout the past few years. However, there is an acute lack of overview, categorization and comparison of such systems. We aim to fill this gap by summarizing and categorizing important methods and techniques for large-scale GNN solutions. In addition, we establish connections between GNN systems, graph processing systems and DL systems.

show abstract

“…GRIP. GRIP [185] points out that there are two modes of computation involved in GCN inference, leading to inefficiency and high latency on existing accelerators. To solve this problem, GRIP first decomposes GCN inference into two parts including the execution of edge-centric and vertex-centric and further designs specialized units to accelerate each part.…”

Section: Graph Learning Acceleratorsmentioning

confidence: 99%

Software Systems Implementation and Domain-Specific Architectures towards Graph Analytics

Jin

Zhao

et al. 2022

Intell Comput

View full text Add to dashboard Cite

Graph analytics, which mainly includes graph processing, graph mining, and graph learning, has become increasingly important in several domains, including social network analysis, bioinformatics, and machine learning. However, graph analytics applications suffer from poor locality, limited bandwidth, and low parallelism owing to the irregular sparse structure, explosive growth, and dependencies of graph data. To address those challenges, several programming models, execution modes, and messaging strategies are proposed to improve the utilization of traditional hardware and performance. In recent years, novel computing and memory devices have emerged, e.g., HMCs, HBM, and ReRAM, providing massive bandwidth and parallelism resources, making it possible to address bottlenecks in graph applications. To facilitate understanding of the graph analytics domain, our study summarizes and categorizes current software systems implementation and domain-specific architectures. Finally, we discuss the future challenges of graph analytics.

show abstract

GRIP: A Graph Neural Network Accelerator Architecture

Cited by 32 publications

References 17 publications

Slice-and-Forge: Making Better Use of Caches for Graph Convolutional Network Accelerators

Slice-and-Forge: Making Better Use of Caches for Graph Convolutional Network Accelerators

The Evolution of Distributed Systems for Graph Neural Networks and Their Origin in Graph Processing and Deep Learning: A Survey

Software Systems Implementation and Domain-Specific Architectures towards Graph Analytics

Contact Info

Product

Resources

About