Recherche - Valorisation
GPU-Accelerated real-time Visualization and Interaction for Coupled Fluid Dynamics.
Pré-Print du CMLA 2013-10 - version du 22 juin 2013
Documents à télécharger :
- CMLA2013-10.pdf.pdf (2041 Ko)
Auteurs : Florian De Vuyst, Christophe Labourdette, Christian Rey
Abstract
For real-time applications (dynamic data-driven applications systems like computer-assisted surgery, command and control, etc.), it is necessary to design fast or strongly-accelerated computational approaches. Reduced-order modeling (ROM) is a candidate methodology that summarizes all the parameter-dependent PDE solutions into an easy-to-compute condensed form. ROM
usually requires an oine learning process that returns the essential components of the solutions.
However, it is known that ROM methodology is not suitable for all problems, especially problems with a large Kolmogorov n-width, like for example dynamical problems involving a continuous multiscale spectrum (like turbulence). In this case, direct simulation is needed and one has to nd acceleration strategies. Graphics Processing units (GPU) are a cheap but relevant way to parallelize computations on thousands of cores leading to speedups of order 200 for some algorithms.
This paper talks about real-time CFD computations allowing for real time visualization and owe interaction.
For real-time applications (dynamic data-driven applications systems like computer-assisted surgery, command and control, etc.), it is necessary to design fast or strongly-accelerated computational approaches. Reduced-order modeling (ROM) is a candidate methodology that summarizes all the parameter-dependent PDE solutions into an easy-to-compute condensed form. ROM
usually requires an oine learning process that returns the essential components of the solutions.
However, it is known that ROM methodology is not suitable for all problems, especially problems with a large Kolmogorov n-width, like for example dynamical problems involving a continuous multiscale spectrum (like turbulence). In this case, direct simulation is needed and one has to nd acceleration strategies. Graphics Processing units (GPU) are a cheap but relevant way to parallelize computations on thousands of cores leading to speedups of order 200 for some algorithms.
This paper talks about real-time CFD computations allowing for real time visualization and owe interaction.
- Type :
- Publication