Investigation of Parallel Programmability and Performance of a Calxeda ARM Server Using OpenCL.

This paper explores the parallel programmability, performance, and energy efficiency of a recently available Calxeda ARM-based server as a potential energy-efficient platform for computationally intensive applications. A novel OpenCL-based parallel programming model for the Calxeda ARM server is achieved via the use of a higher level STandarD Compute Layer (STDCL) application programming interface and a remote procedure call (RPC) implementation. Empirical measurements of the performance of the platform are obtained and presented using an N-body code executed in various configurations. Furthermore, an auto-tuning technique was developed and analyzed for optimization of the N-body algorithm on a specific architecture.