Microbenchmark Based Performance Evaluation of GPU Rendering

Due to advancement in technology and affordable cost effective solution for mobile, smart phones are used widely. Smart phones are able to support multimedia application with the help of graphics processing unit (GPU). The use of GPU for various tasks like drawing geometry based screensaver, texture and lightning effect to produce rendering images is well explored. Speeding up image and vision related applications in embedded devices, the parallelism is exploited in the GPU. Traditional GPU architectures are well suited for the desktop computers, though not efficient for embedded devices. Different GPU vendors use different graphics pipeline architectures. In this paper, we present study of various GPU architectures and their performance evaluation. We also present methodology to evaluate GPU performance based on micro-benchmark accounting for system variables. The test results include the comparison of performance using frame per second and frame time.