Scalable relativistic fluid dynamics for heterogeneous computing

A shift is underway in high performance computing towards new computer architectures that combine traditional CPUs with floating point accelerators, such as GPUs. We have developed a new relativistic fluid code that runs on NVIDIA GPUs using the piecewise-parabolic method, a standard method for compressible fluids found in astrophysics and engineering applications. We present a test study with relativistic magnetohydrodynamics showing that our code scales well in scaling tests with hundreds of nodes.