XEM: Tensor accelerator for AB21 supercomputing artificial intelligence processor

IF 1.3 4区计算机科学 Q3 ENGINEERING, ELECTRICAL & ELECTRONIC

ETRI Journal Pub Date : 2024-10-12 DOI:10.4218/etrij.2024-0141

Won Jeon, Mi Young Lee, Joo Hyun Lee, Chun-Gi Lyuh

{"title":"XEM: Tensor accelerator for AB21 supercomputing artificial intelligence processor","authors":"Won Jeon, Mi Young Lee, Joo Hyun Lee, Chun-Gi Lyuh","doi":"10.4218/etrij.2024-0141","DOIUrl":null,"url":null,"abstract":"<p>As computing systems become increasingly larger, high-performance computing (HPC) is gaining importance. In particular, as hyperscale artificial intelligence (AI) applications, such as large language models emerge, HPC has become important even in the field of AI. Important operations in hyperscale AI and HPC are mainly linear algebraic operations based on tensors. An AB21 supercomputing AI processor has been proposed to accelerate such applications. This study proposes a XEM accelerator to accelerate linear algebraic operations in an AB21 processor effectively. The XEM accelerator has outer product-based parallel floating-point units that can efficiently process tensor operations. We provide hardware details of the XEM architecture and introduce new instructions for controlling the XEM accelerator. Additionally, hardware characteristic analyses based on chip fabrication and simulator-based functional verification are conducted. In the future, the performance and functionalities of the XEM accelerator will be verified using an AB21 processor.</p>","PeriodicalId":11901,"journal":{"name":"ETRI Journal","volume":"46 5","pages":"839-850"},"PeriodicalIF":1.3000,"publicationDate":"2024-10-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.4218/etrij.2024-0141","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"ETRI Journal","FirstCategoryId":"94","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.4218/etrij.2024-0141","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}

引用次数: 0

Abstract

As computing systems become increasingly larger, high-performance computing (HPC) is gaining importance. In particular, as hyperscale artificial intelligence (AI) applications, such as large language models emerge, HPC has become important even in the field of AI. Important operations in hyperscale AI and HPC are mainly linear algebraic operations based on tensors. An AB21 supercomputing AI processor has been proposed to accelerate such applications. This study proposes a XEM accelerator to accelerate linear algebraic operations in an AB21 processor effectively. The XEM accelerator has outer product-based parallel floating-point units that can efficiently process tensor operations. We provide hardware details of the XEM architecture and introduce new instructions for controlling the XEM accelerator. Additionally, hardware characteristic analyses based on chip fabrication and simulator-based functional verification are conducted. In the future, the performance and functionalities of the XEM accelerator will be verified using an AB21 processor.

Abstract Image

查看原文本刊更多论文

XEM：用于 AB21 超级计算人工智能处理器的张量加速器

随着计算系统变得越来越大，高性能计算（HPC）的重要性也与日俱增。特别是随着超大规模人工智能（AI）应用（如大型语言模型）的出现，高性能计算甚至在人工智能领域也变得非常重要。超大规模人工智能和 HPC 中的重要运算主要是基于张量的线性代数运算。为加速此类应用，有人提出了一种 AB21 超级计算人工智能处理器。本研究提出了一种 XEM 加速器，以有效加速 AB21 处理器中的线性代数运算。XEM 加速器具有基于外积的并行浮点运算单元，可高效处理张量运算。我们提供了 XEM 架构的硬件细节，并介绍了用于控制 XEM 加速器的新指令。此外，我们还进行了基于芯片制造的硬件特性分析和基于模拟器的功能验证。未来，我们将使用 AB21 处理器验证 XEM 加速器的性能和功能。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

ETRI Journal 工程技术-电信学

CiteScore

4.00

自引率

7.10%

发文量

审稿时长

6.9 months

期刊介绍： ETRI Journal is an international, peer-reviewed multidisciplinary journal published bimonthly in English. The main focus of the journal is to provide an open forum to exchange innovative ideas and technology in the fields of information, telecommunications, and electronics. Key topics of interest include high-performance computing, big data analytics, cloud computing, multimedia technology, communication networks and services, wireless communications and mobile computing, material and component technology, as well as security. With an international editorial committee and experts from around the world as reviewers, ETRI Journal publishes high-quality research papers on the latest and best developments from the global community.