THINS project: Intermediary report on the SPH code parallelization and validation for free surface
Caricamento...
Data
2011-08
Autori
Massidda, Luca
Titolo del periodico
ISSN
Titolo del volume
Editore
Abstract
In this preliminary report we have presented the performance results of the parallelization of the Armando SPH code on the GPU platform. The results are very encouraging, since a factor of more than 60 on the performance of a CPU core is obtained using an NVIDIA Tesla Card. It appears that GPU acceleration is perfectly suited for the method and on the basis of the profiling analysis it appears that further margin of improvement is possible through a more efficient use of the shared
memory of the card and a reduction in the register usage, to maximize the operational throughput of the computational kernels. In the final report, the code will be further optimized and applied to more complex problems of liquid metal flow and target design.
Descrizione
Keywords
CFD , SPH , smoothed particle hydrodynamics (SPH)