Place of publication: United States, Ann Arbor; ISBN=978-1-369-45321-8
یادداشتهای مربوط به پایان نامه ها
جزئيات پايان نامه و نوع درجه آن
M.S.
نظم درجات
Computer Science
کسي که مدرک را اعطا کرده
Tennessee Technological University
امتياز متن
2016
یادداشتهای مربوط به خلاصه یا چکیده
متن يادداشت
Using Graphic Processing Units (GPU) in general purpose computing is becoming more popular because GPUs provide high throughput and are cost effective. Each new generation of GPUs has vastly increased GPU processing power. The most recent generations of GPUs can execute multiple kernels concurrently in a single GPU. Running concurrent kernels can lead to a significant speedup over a single kernel. However, running multiple kernels at the same time requires proper scheduling to ensure that the kernels take full advantage of the available resources. Previous research works show that some kernels run better together than some other kernels. This thesis presents a framework for experimentation and analysis of concurrently running kernels and presents a reference implementation of the framework. The framework includes guidelines for the functionality of each of the framework tools. This thesis also describes experimental results that show how the reference implementation that is derived from the framework can assist in finding good co-schedules of kernels. In the experiments, good co-schedules exhibit a speedup over serial schedules of as much as 37%.