Performance Measurement of Applications with GPU Acceleration using CUDA
Shangkar Mayanglambam, Allen D. Malony, Matthew J. Sottile · Advances in parallel computing · 2010
Multi-core accelerators offer significant potential to improve the performance of parallel applications. However, tools to help the parallel application developer understand accelerator performance and its impact are scarce. An approach is presented to measure the performance of GPU computations programmed using CUDA and integrate this information with application performance data captured with the TAU Performance System. Test examples are shown to validate the measurement methods. Results for a case study of the GPU-accelerated NAMD molecular dynamics application application are given.