Low-cost and low-power video filtering with parallel many cores

This report examines the advantages of low-cost many-core architectures over graphical processing units (GPU) for massively parallel applications. We show that graphical processing units (GPU) could be costly and power inefficient for workloads such as spatial domain video filtering tasks that take advantage of parallel computing. In this study, we analyze a simple video filtering application on a low-cost parallel programming platform called "Parallella" which includes a Xilinx Zynq all programmable system-on-chip (SoC) along with a parallel processing architecture named "Epiphany" co-processor which includes 16 small RISC cores. We compare it to other hardware and software alternatives in terms of cost, power, and performance. We show that it is possible to use such a low-cost platform for these kinds of workloads without too much programming effort.