Performance Analysis of CUDA and OpenCL By Implementation of Cryptographic Algorithms
Department of Computer Science and Engineering, National Institute of Technology Rourkela, Rourkela-769 008, Odisha, India
National Institute of Technology Rourkela, 2015
@phdthesis{mahapatra2015performance,
title={Performance Analysis of CUDA and OpenCL by Implementation of Cryptographic Algorithms},
author={Mahapatra, Manas},
year={2015}
}
This paper presents a Performance Analysis of CUDA and OpenCL. Three different cryptographic algorithms, i.e. DES, MD5, and SHA-1 have been selected as the benchmarks for extensive analysis of the performance gaps between the two. Our results show that, on the average scenario, CUDA performs 27% better than OpenCL while in the best case scenario it takes over OpenCL by 30%.We also infer that CUDA is more stable and completely masks the access latencies to the shared memory due to the contention of 16 read ports. As far as the optimal number of threads per block goes, 256 threads per block is the most performant choice, proving that the CUDA architecture is able to deal with an increased pressure on the register file without problems as CUDA scores 4.5times over OpenCL in terms of stability.
November 23, 2016 by hgpu
Recent source codes
SYCL-BLAS: An implementation of BLAS using the SYCL open standard for acceleration on OpenCL devices
