Simulation of a multi-core GPU architecture

Abstract

Simulating a GPU architecture under development enables the designer to adjust the architecture and quickly get feedback on performance and efficiency. It is thus possible to explore the solution space more thoroughly and find a better architecture. In this project I improve an existing GPU simulator by debugging it and implementing better logging, and use it to evaluate an experimental multi-core GPU architecture for handheld devices. Simulation was effective in finding problems and bottlenecks in the architecture and pointed out a few changes which will give both better performance and efficiency.