GPU computing for patient-specific model of pulmonary airflow

Takami Yamaguchi, Y. Imai, T. Miki, Takuji Ishikawa

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We propose an implementation of Lattice Boltzmann (LB) method on GPUs for simulating airflow in pulmonary airways with complex branches. An adaptive meshing method is developed for optimizing memory accessing, where the global domain comprises unstructured subdomains, while the local subdomain consists of a structured grids.We also develop a multi-GPU computing method based on a domain decomposition. For strong scaling tests with a subject-specific geometry (12 million LB nodes), the performance on 8 GPUs is approximately 200 GFLOPS, which is 100 times faster computation than 8 CPU cores.

Original languageEnglish
Title of host publicationComputational Modelling of Objects Represented in Images
Subtitle of host publicationFundamentals, Methods and Applications III - Proceedings of the International Symposium, CompIMAGE 2012
Pages239-242
Number of pages4
Publication statusPublished - 2012 Oct 26
Event3rd International Symposium on Computational Modelling of Objects Represented in Images: Fundamentals, Methods and Applications, CompIMAGE 2012 - Rome, Italy
Duration: 2012 Sep 52012 Sep 7

Other

Other3rd International Symposium on Computational Modelling of Objects Represented in Images: Fundamentals, Methods and Applications, CompIMAGE 2012
CountryItaly
CityRome
Period12/9/512/9/7

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Modelling and Simulation

Fingerprint Dive into the research topics of 'GPU computing for patient-specific model of pulmonary airflow'. Together they form a unique fingerprint.

Cite this