The GPU Reconceptualized
Using a GPU as a general-purpose processor is an attractive idea for problems that fit the GPU’s model of parallelism. However many problems do not. In my project I will be conducting a study to investigate if I it is possible to use a multi-core design with less exotic modes of parallelism compared to the GPU, in implementing a graphics pipeline. This will result in a more flexible architecture that can perform both general purpose processing, which is currently performed well by a CPU, as well as problems that are more efficient if executed in parallel.
I will be testing the performance of the proposed architecture against a set of well known general problems. I will then compare the performance of the architecture against a computer with modern computing speed. To test the parallel computation of the architecture I will be simulating the graphics rendering problem. If the system is able to keep up with the GPU in the graphics rendering problem, The system will be proved to be able to implement parallelism.
To test the success of the proposed architecture, it should be able to implement both general computation as well as parallel computation efficiently.
Supervisor: Philip Machanick