Research On General Purpose GPU Computing Technology In The High Performance Computing Platform

In recent years, with the continuous improvement of chip technology and the production levels, GPU has gradually entered the research field of high performance computing for its efficient computing power. With the computing power and programmability of GPU increasing continuously, general purpose computing on GPU (GPGPU) is gradually becoming a research hotspot. Usually the computing with GPGPU utilizes a heterogeneous mode of GPU+CPU, GPU+CPU composite architecture computing cluster has significant differences with traditional computing environments in hardware architectures and software environment design. For the computing center which undertakes large-scale computing tasks, how to make efficient GPU computing technology integrate with its own computing research has been a very essential research topic.The thesis carried out some research work on composite architecture computing cluster which consists of two mainstream GPU of NVIDIA and ATI in Gansu Computing Center. Some contents include some issues, like hardware architecture method of the composite the architecture cluster, cluster batch job automatically processed, analysis and comparison of the computational efficiency on the existing application areas, GPU rendering, etc. We also did some work on computing system upgrade program design, computational efficiency test and comparative summary, and so on. No doubt, all the key research aspects above not only have proposed a solution to existing scientific applications in the computing center, but also have carefully prepared for the design and realization of larger scale GPU+CPU composite architecture cluster and reference and technical support in Gansu Province in near future.