WorkshopsApplication 4: Development of a Cross-Platform Heterogeneous CFD Package Using AVX, CUDA and Qt
79
reads
Modern Computational Fluid Dynamics (CFD) has evolved significantly over the last few years given the introduction of several new powerful computing tools. With the introduction of Graphics Processing Units (GPU) and Advanced Vector eXtensions (AVX), several Finite Volume Methods (FVM) employing vector-split flux computations have enjoyed substantial increases in parallel performance due to the suitability for local (i.e. SIMD) style parallelization. In response to this success, the development of several such schemes has lead to commercialization which has introduced several new challenges. Presented during this talk are the details of such challenges as integrating a GPU/AVX powered CFD solver, using Nvidias CUDA and Intels AVX extensions respectively, with Qt, a popular C-based tool for developing cross platform applications. Examples of such challenges are integrating the compute kernels, graphics and networking engines on platforms such as Linux, Windows and mobile devices. A real-time demonstration using a tablet will be shown during the presentation, and performance characteristics will also be discussed.
reads
Matthew Smith
2013-03-28
11:10:00 - 11:35:00
101 , Mathematics Research Center Building (ori. New Math. Bldg.)
Modern Computational Fluid Dynamics (CFD) has evolved significantly over the last few years given the introduction of several new powerful computing tools. With the introduction of Graphics Processing Units (GPU) and Advanced Vector eXtensions (AVX), several Finite Volume Methods (FVM) employing vector-split flux computations have enjoyed substantial increases in parallel performance due to the suitability for local (i.e. SIMD) style parallelization. In response to this success, the development of several such schemes has lead to commercialization which has introduced several new challenges. Presented during this talk are the details of such challenges as integrating a GPU/AVX powered CFD solver, using Nvidias CUDA and Intels AVX extensions respectively, with Qt, a popular C-based tool for developing cross platform applications. Examples of such challenges are integrating the compute kernels, graphics and networking engines on platforms such as Linux, Windows and mobile devices. A real-time demonstration using a tablet will be shown during the presentation, and performance characteristics will also be discussed.