Sheikh Tousif Rahman

• MEng (Hons) Electrical and Electronic Engineering with Industrial Project
• How Well Can Low-Power Embedded GPUs Handle Large Non-Graphical Tasks?

(A study into low power embedded GPU and CPU run time and power consumption processing non graphical loads through OpenCL framework)

In recent years the shift of the role of the GPU (graphics processing unit), traditionally a fixed function special purpose graphics processor, to a more general purpose programmable unit has meant greater productivity from heterogeneous computing. Heterogeneous computing involves the use of multi-cores, CPUs, GPUs and DSPs (Digital Signal Processors) to synergistically accelerate large computations.

My research scholarship project was centred around trying to learn OpenCL framework to appreciate how this programming model can be implemented across heterogenous units. To evaluate this learning an experiment was devised to identify the productivity of low power GPUs against CPUs when processing non graphical loads. While many research papers have looked into the GPGPU based problems with the goal of maximising speed-up on high end GPUs the goal of this experiment was to give a low power CPU and GPU a large processing task and monitor the power consumption and run times.

Funded by: Newcastle University

Project Supervisor: Dr Rishad Shafik