arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2409.12994cs.ARcs.AIcs.DCcs.LGcs.PF
收藏

性能与功耗:使用 CARAML 对加速器上的 AI 工作负载进行系统评估

Performance and Power: Systematic Evaluation of AI Workloads on Accelerators with CARAML

Jülich Supercomputing Centre(于利希超级计算中心);Forschungszentrum Jülich(于利希研究中心)

Chelsea Maria John, Stepan Nassyr, Carolin Penke, Andreas Herten

AI总结:

本文提出 CARAML 基准测试套件及 jpwr 功耗测量工具,用于系统评估多种硬件加速器上大语言模型与视觉模型训练的性能和能耗。

摘要:

机器学习(ML)技术的快速发展推动了专用硬件加速器的研发,这些加速器旨在实现更高效的模型训练。本文介绍了 CARAML 基准测试套件,该套件用于评估在一系列硬件加速器(包括来自 NVIDIA、AMD 和 Graphcore 的系统)上训练基于 transformer 的大语言模型和计算机视觉模型时的性能与能耗。CARAML 提供了一个紧凑、自动化、可扩展且可复现的框架,用于评估各种新型硬件架构上 ML 工作负载的性能和能耗。文中还详细讨论了 CARAML 的设计与实现,以及一个名为 jpwr 的定制功耗测量工具。

英文摘要:

The rapid advancement of machine learning (ML) technologies has driven the development of specialized hardware accelerators designed to facilitate more efficient model training. This paper introduces the CARAML benchmark suite, which is employed to assess performance and energy consumption during the training of transformer-based large language models and computer vision models on a range of hardware accelerators, including systems from NVIDIA, AMD, and Graphcore. CARAML provides a compact, automated, extensible, and reproducible framework for assessing the performance and energy of ML workloads across various novel hardware architectures. The design and implementation of CARAML, along with a custom power measurement tool called jpwr, are discussed in detail.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑