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 2408.11997cs.AR
收藏

面向低成本矩阵引擎的近似归一化浮点乘加

Floating-Point Multiply-Add with Approximate Normalization for Low-Cost Matrix Engines

Kosmas Alexandridis, Christodoulos Peltekis, Dionysios Filippas, Giorgos Dimitrakopoulos

AI总结:

本文提出在浮点乘加单元中利用近似归一化来降低矩阵引擎硬件复杂度并提升能效,在不牺牲模型精度的前提下,使Bfloat16格式面积和功耗平均降低约16%和13%,且transformer模型精度误差仅1%。

摘要:

机器学习算法的广泛应用需要硬件加速以确保高效性能。这种加速依赖于基于全精度或低精度浮点运算的定制矩阵引擎。然而,传统浮点实现可能功耗很高。本文提出一种方法,以提升用于机器学习算法加速的矩阵引擎的能效。我们的方法在浮点乘加单元中利用近似归一化来降低其硬件复杂度,且不牺牲整体机器学习模型精度。硬件综合结果表明,对于Bfloat16格式,该技术平均将面积和功耗分别降低约16%和13%。此外,在所提方法最高效的配置下,引入transformer模型精度的误差平均为1%。

英文摘要:

The widespread adoption of machine learning algorithms necessitates hardware acceleration to ensure efficient performance. This acceleration relies on custom matrix engines that operate on full or reduced-precision floating-point arithmetic. However, conventional floating-point implementations can be power hungry. This paper proposes a method to improve the energy efficiency of the matrix engines used in machine learning algorithm acceleration. Our approach leverages approximate normalization within the floating-point multiply-add units as a means to reduce their hardware complexity, without sacrificing overall machine-learning model accuracy. Hardware synthesis results show that this technique reduces area and power consumption roughly by 16% and 13% on average for Bfloat16 format. Also, the error introduced in transformer model accuracy is 1% on average, for the most efficient configuration of the proposed approach.

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。

↑