high performance computing on graphics processing units: hgpu.org

Direct numerical simulation (DNS) of turbulence is computationally intensive and typically relies on some form of parallel processing. Spectral kernels used for spatial discretization are a common computational bottleneck on distributed memory architectures. One way to increase DNS algorithms’ efficiency is to parallelize spectral kernels using tightly coupled single-program, multiple-data (SPMD) multiprocessor units with minimal interprocessor communication latency. The authors present techniques to map DNS computations to modern graphics processing units (GPUs), which are characterized by very high memory bandwidth and hundreds of SPMD processors. The article compares the performance between the authors’ parallel algorithm running on a GPU versus the associated CPU implementation of a solver for one of the fundamental nonlinear models of turbulence theory. They also demonstrate a prototype of a scalable computational steering framework based on turbulence simulation and visualization coupling on the GPU.