moDNN: Memory Optimal Deep Neural Network Training on Graphics Processing Units

doi:10.1109/TPDS.2018.2866582

	moDNN: Memory Optimal Deep Neural Network Training on Graphics Processing Units
	Chen, Xiaoming 1; Chen, Danny Ziyi 2; Han, Yinhe 1; Hu, Xiaobo Sharon 2
	2019-03-01
发表期刊	IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
ISSN	1045-9219
卷号	30 期号:3 页码:646-661
摘要	Graphics processing units (GPUs) have been widely adopted to accelerate the training of deep neural networks (DNNs). Although the computational performance of GPUs has been improving steadily, the memory size of modern GPUs is still quite limited, which restricts the sizes of DNNs that can be trained on GPUs, and hence raises serious challenges. This paper introduces a framework, referred to as moDNN (memory optimal DNN training on GPUs), to optimize the memory usage in DNN training. moDNN supports automatic tuning of DNN training code to match any given memory budget (not smaller than the theoretical lower bound). By taking full advantage of overlapping computations and data transfers, we develop new heuristics to judiciously schedule data offloading and prefetching transfers, together with convolution algorithm selection, to optimize memory usage. We further devise a new sub-batch size selection method which also greatly reduces memory usage. moDNN can save memory usage up to 59x, compared with an ideal case which assumes that the GPU memory is sufficient to hold all data. When executing moDNN on a GPU with 12 GB memory, the training time is increased by only 3 percent, which is much shorter than that incurred by the best known approach, vDNN. Furthermore, we propose an optimization strategy for moDNN on multiple GPUs again by utilizing the idea of overlapping data transfers and GPU computations. The results show that 3.7x speedup is attained on four GPUs.
关键词	Deep neural networks graphics processing units memory usage
DOI	10.1109/TPDS.2018.2866582
收录类别	SCI
语种	英语
资助项目	National Science Foundation (NSF)[CCF-1217906] ; National Science Foundation (NSF)[CNS-1629914] ; National Science Foundation (NSF)[CCF-1617735] ; National Science Foundation (NSF)[CCF-1640081] ; Nanoelectronics Research Corporation (NERC) of the Semiconductor Research Corporation (SRC), through Extremely Energy Efficient Collective Electronics (EXCEL), an SRC-NRI Nanoelectronics Research Initiative[2698.004] ; Nanoelectronics Research Corporation (NERC) of the Semiconductor Research Corporation (SRC), through Extremely Energy Efficient Collective Electronics (EXCEL), an SRC-NRI Nanoelectronics Research Initiative[2698.005]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Theory & Methods ; Engineering, Electrical & Electronic
WOS记录号	WOS:000458820700012
出版者	IEEE COMPUTER SOC
引用统计	被引频次：9[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/3413
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Chen, Xiaoming
作者单位	1.Chinese Acad Sci, Inst Comp Technol, State Key Lab Comp Architecture, Beijing 100190, Peoples R China 2.Univ Notre Dame, Dept Comp Sci & Engn, Notre Dame, IN 46556 USA
推荐引用方式 GB/T 7714	Chen, Xiaoming,Chen, Danny Ziyi,Han, Yinhe,et al. moDNN: Memory Optimal Deep Neural Network Training on Graphics Processing Units[J]. IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,2019,30(3):646-661.
APA	Chen, Xiaoming,Chen, Danny Ziyi,Han, Yinhe,&Hu, Xiaobo Sharon.(2019).moDNN: Memory Optimal Deep Neural Network Training on Graphics Processing Units.IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,30(3),646-661.
MLA	Chen, Xiaoming,et al."moDNN: Memory Optimal Deep Neural Network Training on Graphics Processing Units".IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 30.3(2019):646-661.