Abstract
In this paper we demonstrate that it is possible to obtain considerable improvement of performance and energy aware metrics for training of deep neural networks using a modern parallel multi-GPU system, by enforcing selected, non-default power caps on the GPUs. We measure the power and energy consumption of the whole node using a professional, certified hardware power meter. For a high performance workstation with 8 GPUs, we were able to find non-default GPU power cap settings within the range of 160–200 W to improve the difference between percentage energy gain and performance loss by over 15.0%, EDP (Abbreviations and terms used are described in main text.) by over 17.3%, EDS with k = 1.5 by over 2.2%, EDS with k = 2.0 by over 7.5% and pure energy by over 25%, compared to the default power cap setting of 260 W per GPU. These findings demonstrate the potential of today’s CPU+GPU systems for configuration improvement in the context of performance-energy consumption metrics.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Power capping is a mechanism allowing limiting the power draw of a computing device such as a CPU or a GPU, available through Intel RAPL for Intel CPUs and NVIDIA NVML for NVIDIA GPUs, resulting in potentially lower performance but potential for optimization of energy consumption, even throughout extended application execution time [9,10,11].
- 2.
- 3.
Due to space constraints this data is available at https://cdn.files.pg.edu.pl/eti/KASK/RAW2023-paper-supplementary-data/Supplementary_data_Performance_and_power_analysis_of_training_and_performance_quality.pdf.
References
Chen, G., Wang, X.: Performance optimization of machine learning inference under latency and server power constraints. In: 2022 IEEE 42nd International Conference on Distributed Computing Systems (ICDCS), pp. 325–335 (2022). https://doi.org/10.1109/ICDCS54860.2022.00039
Chollet, F.: Xception: deep learning with depthwise separable convolutions. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1800–1807, July 2017. https://doi.org/10.1109/CVPR.2017.195
Czarnul, P., Proficz, J., Drypczewski, K.: Survey of methodologies, approaches, and challenges in parallel programming using high-performance computing systems. Sci. Program. 2020, 4176794:1–4176794:19 (2020). https://doi.org/10.1155/2020/4176794
García-Martín, E., Rodrigues, C.F., Riley, G., Grahn, H.: Estimation of energy consumption in machine learning. J. Parallel Distrib. Comput. 134, 75–88 (2019). https://doi.org/10.1016/j.jpdc.2019.07.007, https://www.sciencedirect.com/science/article/pii/S0743731518308773
He, X., et al.: Enabling energy-efficient DNN training on hybrid GPU-FPGA accelerators. In: Proceedings of the ACM International Conference on Supercomputing, ICS 2021, pp. 227–241. Association for Computing Machinery, New York, NY, USA (2021). https://doi.org/10.1145/3447818.3460371
Jabłońska, K., Czarnul, P.: Benchmarking deep neural network training using multi- and many-core processors. In: Saeed, K., Dvorský, J. (eds.) CISIM 2020. LNCS, vol. 12133, pp. 230–242. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-47679-3_20
Kang, D.K., Lee, K.B., Kim, Y.C.: Cost efficient GPU cluster management for training and inference of deep learning. Energies 15(2), 474 (2022). https://doi.org/10.3390/en15020474, https://www.mdpi.com/1996-1073/15/2/474
Kocot, B., Czarnul, P., Proficz, J.: Energy-aware scheduling for high-performance computing systems: a survey. Energies 16(2), 890 (2023). https://doi.org/10.3390/en16020890, https://www.mdpi.com/1996-1073/16/2/890
Krzywaniak, A., Czarnul, P.: Performance/Energy aware optimization of parallel applications on GPUs under power capping. In: Wyrzykowski, R., Deelman, E., Dongarra, J., Karczewski, K. (eds.) PPAM 2019. LNCS, vol. 12044, pp. 123–133. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-43222-5_11
Krzywaniak, A., Czarnul, P., Proficz, J.: GPU power capping for energy-performance trade-offs in training of deep convolutional neural networks for image recognition. In: Groen, D., de Mulatier, C., Paszynski, M., Krzhizhanovskaya, V.V., Dongarra, J.J., Sloot, P.M.A. (eds.) Computational Science - ICCS 2022, pp. 667–681. Springer, Cham (2022). https://doi.org/10.1007/978-3-031-08751-6_48
Krzywaniak, A., Czarnul, P., Proficz, J.: DEPO: a dynamic energy-performance optimizer tool for automatic power capping for energy efficient high-performance computing. Softw. Pract. Exp. 52(12), 2598–2634 (2022). https://doi.org/10.1002/spe.3139, https://onlinelibrary.wiley.com/doi/abs/10.1002/spe.3139
Lai, C., Ahmad, S., Dubinsky, D., Maver, C.: AI is harming our planet: addressing AI’s staggering energy cost, May 2022. https://www.numenta.com/blog/2022/05/24/ai-is-harming-our-planet/
Leng, J., et al.: GPUWattch: enabling energy optimizations in GPGPUs. SIGARCH Comput. Archit. News 41(3), 487–498 (2013). https://doi.org/10.1145/2508148.2485964
Mazuecos Pérez, M.D., Seiler, N.G., Bederián, C.S., Wolovick, N., Vega, A.J.: Power efficiency analysis of a deep learning workload on an IBM “Minsky’’ Platform. In: Meneses, E., Castro, H., Barrios Hernández, C.J., Ramos-Pollan, R. (eds.) CARLA 2018. CCIS, vol. 979, pp. 255–262. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-16205-4_19
McDonald, J., Li, B., Frey, N., Tiwari, D., Gadepally, V., Samsi, S.: Great power, great responsibility: recommendations for reducing energy for training language models. In: Findings of the Association for Computational Linguistics: NAACL 2022. Association for Computational Linguistics (2022). https://doi.org/10.18653/v1/2022.findings-naacl.151
Rouhani, B.D., Mirhoseini, A., Koushanfar, F.: Delight: adding energy dimension to deep neural networks. In: Proceedings of the 2016 International Symposium on Low Power Electronics and Design, ISLPED 2016, pp. 112–117. Association for Computing Machinery, New York, NY, USA (2016). https://doi.org/10.1145/2934583.2934599
Schuchart, J., et al.: The READEX formalism for automatic tuning for energy efficiency. Computing 99(8), 727–745 (2017)
Tao, Y., Ma, R., Shyu, M.L., Chen, S.C.: Challenges in energy-efficient deep neural network training with FPGA. In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 1602–1611 (2020). https://doi.org/10.1109/CVPRW50498.2020.00208
Wang, F., Zhang, W., Lai, S., Hao, M., Wang, Z.: Dynamic GPU energy optimization for machine learning training workloads. IEEE Trans. Parallel Distrib. Syst. 33(11), 2943–2954 (2022). https://doi.org/10.1109/TPDS.2021.3137867
Xu, Y., Martínez-Fernández, S., Martinez, M., Franch, X.: Energy efficiency of training neural network architectures: an empirical study (2023)
Yang, H., Zhu, Y., Liu, J.: ECC: platform-independent energy-constrained deep neural network compression via a bilinear regression model. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11198–11207. IEEE Computer Society, Los Alamitos, CA, USA, June 2019. https://doi.org/10.1109/CVPR.2019.01146, https://doi.ieeecomputersociety.org/10.1109/CVPR.2019.01146
Yang, T.J., Chen, Y.H., Emer, J., Sze, V.: A method to estimate the energy consumption of deep neural networks. In: 2017 51st Asilomar Conference on Signals, Systems, and Computers, pp. 1916–1920 (2017). https://doi.org/10.1109/ACSSC.2017.8335698
Yang, Z., Meng, L., Chung, J.W., Chowdhury, M.: Chasing low-carbon electricity for practical and sustainable DNN training (2023)
You, J., Chung, J.W., Chowdhury, M.: Zeus: understanding and optimizing GPU energy consumption of DNN training. In: 20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 2023), pp. 119–139. USENIX Association, Boston, MA, April 2023. https://www.usenix.org/conference/nsdi23/presentation/you
Zou, P., Li, A., Barker, K., Ge, R.: Indicator-directed dynamic power management for iterative workloads on GPU-accelerated systems. In: 2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID), pp. 559–568 (2020). https://doi.org/10.1109/CCGrid49817.2020.00-37
Acknowledgment
We would like to thank the administrator of the HPC server at Department of Computer Architecture at the GUT, dr Tomasz Boiński, for support regarding setting up the testbed environment. This work is supported by CERCIRAS COST Action CA19135 funded by the COST Association as well as statutory funds of Dept. of Computer Architecture, Faculty of Electronics, Telecommunications and Informatics, Gdańsk Tech.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2024 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Koszczał, G., Dobrosolski, J., Matuszek, M., Czarnul, P. (2024). Performance and Energy Aware Training of a Deep Neural Network in a Multi-GPU Environment with Power Capping. In: Zeinalipour, D., et al. Euro-Par 2023: Parallel Processing Workshops. Euro-Par 2023. Lecture Notes in Computer Science, vol 14352. Springer, Cham. https://doi.org/10.1007/978-3-031-48803-0_1
Download citation
DOI: https://doi.org/10.1007/978-3-031-48803-0_1
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-48802-3
Online ISBN: 978-3-031-48803-0
eBook Packages: Computer ScienceComputer Science (R0)