Optimization of the Particle-based Volume Rendering for GPUs with Hiding Data Transfer Latency
Graduate School of Humanities and Sciences, Nara Women’s University, Nara, 630-8506, JAPAN
The 2011 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA’11), 2011
@article{nakao2011optimization,
title={Optimization of the Particle-based Volume Rendering for GPUs with Hiding Data Transfer Latency},
author={Nakao, K. and Matsui, E. and Yoshii, N. and Takata, M. and Joe, K.},
year={2011}
}
In this paper, we present the optimization of the particle-based volume rendering for GPU platforms. In general, data transfer between CPU and GPU accompanies long latency. Using page lock memory of the CUDA runtime API, data area is selected so that the data transfer between CPU and GPU becomes faster to reduce the execution time. In the meantime, Using streams, the overlap of data transfer and the execution of kernels is achieved. As the result of experiment with the voxel data of a typhoon (1,188 x 979 x 64, 140MB), data transfer time and kernel execution time are improved and bring out about 30% performance of the GPU.
November 30, 2011 by hgpu
