GPU Progress in Sparse Matrix Solvers for Applications in Computational Mechanics

Current trends in high performance computing (HPC) are advancing towards the use of graphics processing units (GPUs) to achieve speed-ups through the extraction of fine-grain parallelism of applications. Since 2007 NVIDIA GPUs have been developed exclusively for computational tasks as massively-parallel co-processors to the CPU, which provide new opportunities for sparse matrix solvers for applications in computational mechanics. This paper examines the current state of GPU parallel solvers in computational mechanics with a review of applications in computational structural mechanics (CSM) and computational fluid dynamics (CFD) that support product development for the manufacturing industry. Computational efficiency and simulation turnaround times continue to be important factors behind scientific and engineering decisions to develop models at higher fidelity, and recent history has shown that a rapid simulation capability with increased model fidelity has the potential to transform current practices in engineering analysis and design optimization. GPU trends for CSM and CFD software show substantial gains in parallel efficiency from a second-level fine-grain parallelism under first-level distributed memory parallel through implementation of hybrid CPU-GPU co-processing. Examples are provided that compare results with the use of conventional CPUs with and without GPU acceleration.