Abstract
Deep neural networks (DNN) offer great opportunities for solving many problems associated with processing large-scale data. Building and using deep neural networks requires large computational resources. In this regard, the question naturally arises about the possibility of using HPC-systems for the implementation of DNN. In order to better understand the performance implications of DNN on High Performance clusters we analyze the performance of several DNN-models over 2 HPC systems: the Lomonosov-2 supercomputer (the section with processors equipped with P100 GPUs) and the Polus high-performance cluster based on IBM Power8 processors with P100 GPUs. Comparing these frameworks is interesting as they represent different types of processors (Intel for Lomonosov-2 and IBM for Polus). Apart from different processor architectures, these systems feature different internode communications, which may affect the performance of the analysed algorithms in case of parallel and distributed implementation of neural network algorithms. The studies were carried out on the basis of the PyTorch framework.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Specifications of supercomputer Lomonosov-2. https://parallel.ru/cluster/lomonosov2.html
Specifications of cluster Polus. http://hpc.cs.msu.su/polus
Ben-Nun, T., Hoefler, T.: Demystifying parallel and distributed deep learning: an in-depth concurrency analysis. arXiv:1802.09941 (2018)
Liu, L., Wu, Y., Wei, W., Cao, W., Sahin, S., Zhang, Q.: Benchmarking deep learning frameworks: design considerations, metrics and beyond. In: 2018 IEEE 38th International Conference on Distributed Computing Systems (ICDCS), Vienna, pp. 1258–1269 (2018). https://doi.org/10.1109/icdcs.2018.00125
Zhang, X., Wang, Y., Shi, W.: pCAMP: performance comparison of machine learning packages on the edges. In: USENIX Workshop on Hot Topics in Edge Computing (HotEdge 2018) (2018)
Asaadi, H., Chapman, B.: Comparative study of deep learning framework in HPC environments. In: 2017 New York Scientific Data Summit (NYSDS), pp. 1–7 (2017)
Tato, A.G.: Evaluation of machine learning fameworks on finis terrae II. arXiv:1801.04546 (2018)
Mericas, A., et al.: IBM POWER8 performance features and evaluation. IBM J. Res. Dev. 59, 6:1–6:10 (2015)
PyTorch. https://pytorch.org
Sandler, M., Howard, A., Zhu, M., Zhmoginov, A.: MobileNetV2: inverted residuals and linear bottlenecks. In: 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4510–4520 (2018). arXiv:1801.04381
Simonyan, K.: Very deep convolutional networks for large-scale image recognition. arXiv:1409.1556 (2014)
Iandola, F.N., Han, S., Moskewicz, M.W., Ashraf, K., Dally, W.J.: SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5 MB model size. arXiv:1602.07360 (2016)
Kingma, D., Jimmy, B.: Adam: a method for stochastic optimization. In: Contribution to International Conference on Learning Representations, San Diego, 7–9 May 2015 (2015)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Buryak, D. et al. (2019). An Experimental Study of Deep Neural Networks on HPC Clusters. In: Voevodin, V., Sobolev, S. (eds) Supercomputing. RuSCDays 2019. Communications in Computer and Information Science, vol 1129. Springer, Cham. https://doi.org/10.1007/978-3-030-36592-9_40
Download citation
DOI: https://doi.org/10.1007/978-3-030-36592-9_40
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-36591-2
Online ISBN: 978-3-030-36592-9
eBook Packages: Computer ScienceComputer Science (R0)