Graphics Processor performance analysis for 3D applications

The performance analysis of Graphics Processor Unit (GPU) is important for analyzing and fine-tuning current graphics processors as well as for comparing the performance of different technologies. In this paper, we present an analytical model to calculate the total time it takes for a GPU to retire one frame for a given benchmark. The model also estimates the total retirement time for the same frame for different GPU configuration. The model consists of two stages. The first stage entails establishing the measured baseline for a specific frame on a given graphics card, and the second stage entails adjusting the measured baseline and estimating the time it takes to process all draw calls for the same frame on a different GPU. The model considers the impact of pipeline bottlenecks to process a specific frame, estimates the minimum time it takes to process that frame, and re-parameterize the baseline for a different graphics card to calculate new frame retirement times at two different memory frequencies. We propose an analytical estimation method to estimate frame retirement time for a different GPU at higher memory frequencies based on the new adjusted measured baseline.