Accelerated parallel computation of field quantities for the boundary element method applied to stress analysis using multi-core CPUs, GPUs and FPGAs

Computation in engineering and science can often benefit from acceleration due to lengthy calculation times for certain classes of numerical models. This paper, using a practical example drawn from computational mechanics, formulates an accelerated boundary element algorithm that can be run in parallel on multi-core CPUs, GPUs and FPGAs. Although the computation of field quantities, such as displacements and stresses, using boundary elements is specific to mechanics, it can be used to highlight the strengths and weaknesses of using hardware acceleration. After the necessary equations were developed and the algorithmic implementation was summarized, each hardware platform was used to run a set of test cases. Both time-to-solution and relative speedup were used to quantify performance as compared to a serial implementation and to a multi-core implementation as well. Parameters, such as the number of threads in a workgroup and power consumption were considered and recommendations are given concerning the merits of each hardware accelerator.