Bench IP: Benchmarking Intelligence Processors
Bench IP: Benchmarking Intelligence Processors
复制标题
Bench IP：智能处理器基准测试
DOI：
10.1007/s11390-018-1805-8
复制
发表时间：
2018
影响因子：
0.7
通讯作者：
Chen Tian Shi
中科院分区：
文献类型：
--
作者：
Tao Jin Hua;Du Zi Dong;Guo Qi;Lan Hui Ying;Zhang Lei;Zhou Sheng Yuan;Xu Ling Jie;Liu Cong;Liu Hai Feng;Tang Shan;Rush Allen;Chen Willian;Liu Shao Li;Chen Yun Ji;Chen Tian Shi
对深度学习的日益关注极大地刺激了智能处理硬件的设计。各种新兴的智能处理器需要标准的基准来进行公平的比较和系统优化（软件和硬件）。然而，现有的基准是不适合基准的智能处理器，由于其非多样性和非代表性。此外，缺乏标准的基准方法进一步加剧了这一问题。在本文中，我们提出了BenchIP，一个基准套件和基准测试方法的智能处理器。BenchIP中的基准测试套件由两组基准测试组成：微基准测试和宏基准测试。微基准测试由单层网络组成。它们主要用于瓶颈分析和系统优化。宏观基准包含最先进的工业网络，以提供不同平台的现实比较。我们还提出了一个标准的基准测试方法，建立在一个工业软件栈和评估指标，全面反映了评估的智能处理器的各种特性。BenchIP用于评估各种硬件平台，包括CPU，GPU和加速器。BenchIP很快就会开源。
The increasing attention on deep learning has tremendously spurred the design of intelligence processing hardware. The variety of emerging intelligence processors requires standard benchmarks for fair comparison and system optimization (in both software and hardware). However, existing benchmarks are unsuitable for benchmarking intelligence processors due to their non-diversity and nonrepresentativeness. Also, the lack of a standard benchmarking methodology further exacerbates this problem. In this paper, we propose BenchIP, a benchmark suite and benchmarking methodology for intelligence processors. The benchmark suite in BenchIP consists of two sets of benchmarks: microbenchmarks and macrobenchmarks. The microbenchmarks consist of single-layer networks. They are mainly designed for bottleneck analysis and system optimization. The macrobenchmarks contain state-of-the-art industrial networks, so as to offer a realistic comparison of different platforms. We also propose a standard benchmarking methodology built upon an industrial software stack and evaluation metrics that comprehensively reflect various characteristics of the evaluated intelligence processors. BenchIP is utilized for evaluating various hardware platforms, including CPUs, GPUs, and accelerators. BenchIP will be open-sourced soon.