high performance computing on graphics processing units: hgpu.org

The graphics processing unit (GPU) has seen significant increase in performance over the past few years. Hence the interest in using GPUs for more general purposes has increased. The higher number of cores on a GPU allows it to outperform central processing units (CPUs). However, since in certain aspects instructions executed on the GPU must be executed in lock-step, generally the same instruction needs to be executed on multiple data sets. A deshredder program was created to test the performance gain where using the GPU to perform matching of shreds. Tests were performed using varying numbers of shreds to test how the transfer overhead between the CPU and GPU affects the timing, and how accurate a simple matching algorithm able to be executed on a GPU can be. The deshredder program showed a significant speedup when run on the GPU compared to with execution on a CPU, thereby confirming the hypothesis that a deshredding program could benefit from being executed on a GPU.