Transitive Array: An Efficient GEMM Accelerator with Result Reuse
Name
3695053.3731043.pdf
Size
1.23 MB
Format
Adobe PDF
Checksum (MD5)
fc7270c166e9a040cd83cbba06279a9c
Author(s)
Guo, Cong
Wei, Chiyue
Tang, Jiaming
Duan, Bowen
Han, Song
Li, Hai
Chen, Yiran
Date Issued
June 20, 2025
Publisher
ACM|Proceedings of the 52nd Annual International Symposium on Computer Architecture
Citation
Cong Guo, Chiyue Wei, Jiaming Tang, Bowen Duan, Song Han, Hai Li, and Yiran Chen. 2025. Transitive Array: An Efficient GEMM Accelerator with Result Reuse. In Proceedings of the 52nd Annual International Symposium on Computer Architecture (ISCA '25). Association for Computing Machinery, New York, NY, USA, 990–1004.
Version
Final published version
Abstract
Deep Neural Networks (DNNs) and Large Language Models (LLMs) have revolutionized artificial intelligence, yet their deployment faces significant memory and computational challenges, especially in resource-constrained environments. Quantization techniques have mitigated some of these issues by reducing data precision, primarily focusing on General Matrix Multiplication (GEMM). This study introduces a novel sparsity paradigm, transitive sparsity, which leverages the reuse of previously computed results to substantially minimize computational overhead in GEMM operations. By representing transitive relations using a directed acyclic graph, we develop an efficient strategy for determining optimal execution orders, thereby overcoming inherent challenges related to execution dependencies and parallelism. Building on this foundation, we present the Transitive Array, a multiplication-free accelerator designed to exploit transitive sparsity in GEMM. Our architecture effectively balances computational workloads across multiple parallel lanes, ensuring high efficiency and optimal resource utilization. Comprehensive evaluations demonstrate that the Transitive Array achieves approximately 7.46 × and 3.97 × speedup and 2.31 × and 1.65 × energy reduction compared to state-of-the-art accelerators such as Olive and BitVert while maintaining comparable model accuracy on LLaMA models.
Description
ISCA ’25, Tokyo, Japan
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1145/3695053.3731043
