CUTE: A scalable CPU-centric and Ultra-utilized Tensor Engine for convolutions

doi:10.1016/j.sysarc.2024.103106

	CUTE: A scalable CPU-centric and Ultra-utilized Tensor Engine for convolutions
	Li, Wenqing 1,2; Ye, Jinpeng 1,2; Zhang, Fuxin 1,2; Liu, Tianyi 3; Zhang, Tingting 1,4; Wang, Jian 1,2
	2024-04-01
发表期刊	JOURNAL OF SYSTEMS ARCHITECTURE
ISSN	1383-7621
卷号	149 页码:15
摘要	Convolution is a fundamental and computationally expensive primitive and finds ubiquitous in deep neural networks (DNNs). The evolving DNNs have spurred the emergence of numerous accelerators and they successfully achieve high throughput. However, for DNN inference with small batch sizes, the computational resources of the accelerators are often under-utilized, and the overhead of offloading is significant. Compared to accelerators, the CPU can better meet fast response requirements of inference, flexibly handle various models, and is suitable for various scenarios (from edge to data center). Therefore, CPU remains an attractive platform for DNN inference, despite the sub-optimal performance, and resource efficiency. In this paper, we propose CUTE, a scalable CPU-centric and ultra-utilized tensor engine for convolutions. It co-designs data flow and hardware architecture to leverage the data reuse and parallelism of convolutions. CUTE is composed of several small tensor elements (TEs) and two-level buffers. It employs a decoupled accessexecution architecture and greedy strategy to feed data to TEs, enabling it to achieve ultra utilization and great scalability. CUTE is tightly coupled with the CPU to minimize offloading latency, thereby providing efficient convolution computing capabilities for the system. Experimental results show that under the same bandwidth, CUTE achieves an average performance improvement of 3.8x compared with the CPU AVX512 unit and 1.6x compared with the CPU AMX unit. Besides, CUTE achieves a speedup of 7.0x and 3.9x over Nvidia V100 GPU and Eyeriss accelerator respectively, due to higher utilization of computing units.
关键词	Tensor engine Convolution Scalable architecture CPU-centric Utilization
DOI	10.1016/j.sysarc.2024.103106
收录类别	SCI
语种	英语
资助项目	Strategic Priority Research Program of the Chinese Academy of Sciences[XDC05020100]
WOS研究方向	Computer Science
WOS类目	Computer Science, Hardware & Architecture ; Computer Science, Software Engineering
WOS记录号	WOS:001207560600001
出版者	ELSEVIER
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/38703
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Wang, Jian
作者单位	1.Chinese Acad Sci, Inst Comp Technol, State Key Lab Processors, Beijing, Peoples R China 2.Univ Chinese Acad Sci, Beijing, Peoples R China 3.Univ Texas San Antonio, San Antonio, TX USA 4.Loongson Technol Corp Ltd, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Li, Wenqing,Ye, Jinpeng,Zhang, Fuxin,et al. CUTE: A scalable CPU-centric and Ultra-utilized Tensor Engine for convolutions[J]. JOURNAL OF SYSTEMS ARCHITECTURE,2024,149:15.
APA	Li, Wenqing,Ye, Jinpeng,Zhang, Fuxin,Liu, Tianyi,Zhang, Tingting,&Wang, Jian.(2024).CUTE: A scalable CPU-centric and Ultra-utilized Tensor Engine for convolutions.JOURNAL OF SYSTEMS ARCHITECTURE,149,15.
MLA	Li, Wenqing,et al."CUTE: A scalable CPU-centric and Ultra-utilized Tensor Engine for convolutions".JOURNAL OF SYSTEMS ARCHITECTURE 149(2024):15.