Cost-Performance Comparison of Various Accelerator Implementation Platforms for Deep Convolutional Neural Network

  • Yechan Yu
  • HoJin Kim
  • Jinjoo Ha
  • Daewoo Kim
  • Kang YiEmail author
Conference paper
Part of the Communications in Computer and Information Science book series (CCIS, volume 931)


Due to the high accuracy, DCNN is a popular deep learning approach for object recognition and classification. But, the computing complexity of DCNN is too high for real-time application. Therefore, many acceleration methods like GPU and FPGAs are developed and competing with each other. The purpose of this paper is to assess the pros and the cons of many acceleration methods including GPGPU and FPGA-based approaches like Xilinx SDSoC, and Xilinx SDAccel. We will consider the installation cost (board price) as well as the operation cost (the energy consumption) and the speed of each acceleration method in the analysis.


Reconfigurable high-performance computing Heterogeneous computing systems Intelligent computing and neural networks 



This work was supported by National program for Excellence in software at Handong Global University (2017-0-00130) funded by Ministry of Science and ICT in Korea.


  1. 1.
    Farfade, S.S., Saberian, M.J., Li, L.J.: Multi-view face detection using deep convolutional neural networks. In: Proceedings of the 5th ACM on International Conference on Multimedia Retrieval, China, pp. 643–650 (2015)Google Scholar
  2. 2.
    Kim, I.J., Xie, X.: Handwritten Hangul recognition using deep convolutional neural networks. Int. J. Doc. Anal. Recogn. (IJDAR) 18(1), 1–13 (2015)CrossRefGoogle Scholar
  3. 3.
    Park, H., et al.: Optimizing DCNN FPGA accelerator design for handwritten Hangul character recognition: work-in-progress. In: The 2017 International Conference (2017).
  4. 4.
    Yu, Y., Kim, H., Park, Y., Kim, D., Choi, K., Yi, K.: Hardware acceleration using Xilinx SDSoC development environment for real-time number image recognition deep learning system on Zedboard, Korea Multimedia Society (2018). ISSN 1229-778XGoogle Scholar
  5. 5.
    Tompson, J., Schlachter, K.: An Introduction to the OpenCL Programming Model. Pearson Education, London (2012)Google Scholar
  6. 6.
    Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: Advances in Neural Information Processing Systems, pp. 1097–1105 (2012)Google Scholar
  7. 7.
    Sozzo, E.D., Baghdadi, R., Amarasinghe, S.P., Santambrogio, M.D.: A common backend for hardware acceleration on FPGA. In: IEEE International Conference on Computer (2017).
  8. 8.
    Weijema, E.: How to calculate electrical costs for cooling and power consumption (2011).

Copyright information

© Springer Nature Singapore Pte Ltd. 2019

Authors and Affiliations

  • Yechan Yu
    • 1
  • HoJin Kim
    • 1
  • Jinjoo Ha
    • 1
  • Daewoo Kim
    • 1
  • Kang Yi
    • 1
    Email author
  1. 1.School of Computer Science and Electrical EngineeringHandong Global UniversityPohangRepublic of Korea

Personalised recommendations