high performance computing on graphics processing units: hgpu.org

Traditional ways of increasing computer performance has been increasing speed and bit size. Although this kept us going for more than half of a century, methodology has hit a major road block due to power consumption and heat dissipation. The remedy found for this problem has been creating multiple cores/heterogeneous computing systems on a single chip. By introducing multiple cores, efficient memory execution and correct data sharing the performance can be increased without paying power penalty. In this paper, we discuss the implications caused by integrating multiple cores on a single chip and review some of the techniques generated for solving these complications. Among these are techniques for establishing effective communications, data integrity and buffer management.