The Effect of High Performance Computer on Deep Neural Network

Xin Zhang,

Ting Zhang,

Jiang Lu,3*Email

Xingang Fu

Francisco Reveriano

1Department of Computer Engineering, University of Houston-Clear Lake, Houston, USA

2Computer Science & Engineering Technology, University of Houston-Downtown, Houston USA

   3Department of Computer Engineering, University of Houston-Clear Lake, Houston, USA

4Electrical Engineering and Computer Science, Texas A&M University-Kingsville, Kingsville, USA

5Department of Computer Engineering, University of Houston-Clear Lake, Houston, USA

Abstract

In recent years, convolutional neural networks (CNN) have been extended widely to a large number of computer vision applications such as image classification, image detection, image segmentation, etc. In this paper, the three image processing applications are implemented by integrating with CNNs and the High Performance Computing (HPC) systems. To observe the performance of HPC, three CNN models for each image processing application have been trained with different values of parameters and their training times are provided in results. Four computing systems, Google Colaboratory with CPU, Google Colaboratory with GPU, Google Cloud with HPC, and XSEDE with HPC are compared in the work. The training program is suggested to use the parallel algorithm when GPU is available. This project explores that the HPC with GPU has the highest work efficiency regarding operating time.

The Effect of High Performance Computer on Deep Neural Network