GPU Accelerated Radiation-Magneto-Hydrodynamics

如今，天体物理学的研究在很大程度上是基于大规模的数值模拟。最近和正在进行的硬件发展清楚地表明，下一代超级计算机将由多核节点主导。其中，最有希望的经久不衰的成就将建立在图形处理单元(GPU)上，目前图形处理单元(GPU)将500多个计算核心整合到一个GPU上。不幸的是，大多数天体物理软件包还不能利用GPU功能。在这个拟议的项目中，我们将进一步移植我们版本的磁流体动力学程序Flash和辐射传输程序Phoenix，以有效地利用GPU的能力。这些开发将补充最近在汉堡斯特恩沃特安装的GPU集群和计划在汉堡大学雷欣兹特姆地区安装的GPU超级计算机。

Research in astrophysics is nowadays heavily based on large scalenumerical simulations. The recent and ongoing hardware developmentshows clearly that the next generation supercomputer will be dominatedby many-core nodes. Among those, the most promising enduringachievement will be build on graphics processing units (GPUs),which combine these days more than 500 compute cores on a singleGPU. Unfortunately, the most astrophysical software packages are notyet able to utilize the GPU capabilities. In this proposed project wewill further port our version of magneto-hdyrodyamical code FLASH andthe radiation transport code PHOENIX to efficiently utilize thecapabilities of GPUs. These developments will complement the recently installed GPU-cluster at the Hamburger Sternwarte and the planned GPU supercomputerat the Regionale Rechenzentrum of the University of Hamburg.