Performance Analysis of Application Kernels in Multi/Many-Core Architectures

In recent years, advancement in technology and computing led to huge amounts of data being generated. Thus, HighPerformance Computing (HPC) plays an ever growing role in processing these large datasets in a timely fashion. Our analysis consist of few important throughput computing app kernels which have high degree of parallelism and makes them excellent candidates for evaluation on high end multi-core CPUs, and manycore GPUs. In this work, we performed a performance comparison of important app kernels like Image Convolution, Histogram and Bilateral filtering in multi-core CPU, many-core NVIDIA GPUs in addition to comparing our research framework GPU enabled ManyTask Computing (GeMTC). GeMTC is an execution model and runtime system which enables NVIDIA GPUs to be programmed with many concurrent and independent tasks of potentially short or variable duration. In this work we provide a thorough performance analysis between CPU, CUDA, and the GeMTC framework. Through this we better understand the behavior of different applications that belong to the Many-Task Computing paradigm. The results show that the GeMTC framework shows promising results for Many-Task Computing workloads running on NVIDIA GPUs.