high performance computing on graphics processing units: hgpu.org

Field Programmable Gate Arrays (FPGAs) have been widely used for accelerating machine learning algorithms. However, the high design cost and time for implementing FPGA-based accelerators using traditional HDL-based design methodologies has discouraged users from designing FPGA-based accelerators. In recent years, a new CAD tool called Intel FPGA SDK for OpenCL (IFSO) allowed fast and efficient design of FPGA-based hardware accelerators from high level specification such as OpenCL. Even software engineers with basic hardware design knowledge could design FPGA-based accelerators. In this thesis, IFSO has been used to explore acceleration of k-Nearest-Neighbour (kNN) algorithm and Speckle Reducing Anisotropic Diffusion (SRAD) simulation using FPGAs. kNN is a popular algorithm used in machine learning. Bitonic sorting and radix sorting algorithms were used in the kNN algorithm to check if these provide any performance improvements. Acceleration of SRAD simulation was also explored. The experimental results obtained for these algorithms from FPGA-based acceleration were compared with the state of the art CPU implementation. The optimized algorithms were implemented on two different FPGAs (Intel Stratix A7 and Intel Arria 10 GX). Experimental results show that the FPGA-based accelerators provided similar or better execution time (up to 80X) and better power efficiency (75% reduction in power consumption) than traditional platforms such as a workstation based on two Intel Xeon processors E5-2620 Series (each with 6 cores and running at 2.4 GHz).