@inproceedings{b6f8954797544594bc16c8a81a8ece53,
title = "Research on SPH parallel acceleration strategies for multi-GPU platform",
abstract = "This paper proposes an acceleration strategy for SPH on single-node multi-GPU platform. First the acceleration strategy for SPH on single-GPU is studied in conjunction with the characteristics of architecture. Then the changing pattern of SPH's computation time has been discussed. Based on the fact that the changing pattern is rather slow, using a simple dynamic load balancing algorithm an acceptable load balance is obtained on multi-GPU. Finally, an almost linear speedup is achieved on multi-GPU by further optimizing dynamic load balancing algorithm and communication strategy among multiple GPUs.",
keywords = "SPH, communication optimization, dynamic load balancing, multi-GPU",
author = "Lei Hu and Xukun Shen and Xiang Long",
year = "2013",
doi = "10.1007/978-3-642-45293-2\_8",
language = "英语",
isbn = "9783642452925",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "104--118",
booktitle = "Advanced Parallel Processing Technologies - 10th International Symposium, APPT 2013, Revised Selected Papers",
note = "10th International Symposium on Advanced Parallel Processing Technologies, APPT 2013 ; Conference date: 27-08-2013 Through 28-08-2013",
}