Ultra Low-Cost Parallel Computing Using Low-End Phone Processors

Abstract

In the past few years, as the popularity and demand for smart mobile devices has grown, the performance of mobile phone processors has increased tremendously. At the same time, mobile processors have continued to have low costs, consume low power, and have compact designs, making them great processors for manycore, parallel computing. In this thesis, we present an ultra low-cost computer consisting of low-end mobile processors as a cost-effective alternative to high performance computing. We use the Allwinner H5, a quadcore, ARM-based processor that costs $4.50, to design a PCB with eight H5 processors that can connect to other boards in a flexible architecture and scale up to include many cores. Compared to a quadcore consumer processor and a 20-core server processor, the H5 has a better performance-to-cost ratio, demonstrating a peak ratio of 48.9 Mop/s/$, 1.63x the ratio for the Intel Xeon E5-2670 v2 processor and 3.71x the ratio for the Intel Core i7-4870HQ processor.