Effects of GPU and CPU Loads on Performance of CUDA Applications

General purpose computing on GPUs provides a way for certain applications to benefit from a commonly available massively parallel architecture. As such deployment becomes more widespread, multiple GPU applications will have to execute on the same hardware in systems that have only one GPU. The aggregate loads of the GPU and CPU impact the performance of each application. This work investigates the effects of CPU and GPU loads on the performance of two CUDA GPU applications with significantly different CPU-GPU interaction profiles: implementations of the AES encryption and Keccak hashing algorithms. The percentage degradation in performance of these applications from CPU and GPU loads indicates dependence on the total execution time of the application, with the greatest degradation for the shortest execution times. Performance degradations as high as 22% and 36% were observed for CPU and GPU loads, respectively.