Evaluating the maturity of OpenFOAM simulations on GPGPU for bio-fluid applications


Creative Commons License

Duran A. , Pıskın S., Tunçel M.

The Emerging Technology (EMiT) Conference, Barcelona, İspanya, 2 - 03 Haziran 2016, ss.11-14

  • Basıldığı Şehir: Barcelona
  • Basıldığı Ülke: İspanya
  • Sayfa Sayıları: ss.11-14

Özet

It is important to deal with the computational challenges for bio-medical fluid flow simulations and an OpenFOAM solver, icoFoam, for the large matrices coming from the simulation of blood flow in arteries on different HPC clusters. The flow problem produced various matrices as the time advances in simulation. In this study we examined the behaviour of the solvers for ill-conditioned matrices. We compared the CPU performance of the iterative solver icoFoam and the hybrid parallel codes (MPI+OpenMP) of a direct solver SuperLU_DIST 4.0 (see [2]) at TGCC Curie (a Tier-0 system) thin nodes at CEA, France (see [5]). Moreover, we compared the performance of the hybrid parallel codes of MPI+OpenMP+CUDA versus MPI+OpenMP implementation of SuperLU_DIST 4.0 at TGCC Curie (a Tier-0 system) hybrid nodes of CPU + GPU at CEA, France (see [5]). We discuss the performance, scalability and robustness of OpenFOAM on GPGPU cluster. We present our results regarding the speed-up of the solvers for the large matrices of size up to 20 million x 20 million. The authors thank to PRACE, GENCI and CEA for the opportunity to conduct our research in the frame of the Project 2010PA2505 awarded under the 18th Call for PRACE Preparatory Access.