Investigations into techniques to accelerate memory intensive GPGPU applications

Recent advancements in the architecture of Graphic Processing Unit (GPU), enables the acceleration of many general purpose applications. Even with high memory bandwidth, GPUs are still faced with the challenge of accelerating highly memory intensive applications. To overcome this challenge this paper investigates the impact of scaling up of the memory partitions and also scaling of frequency of the memory system on the performance of different types of applications. It is observed that all memory intensive applications experience significant improvement in the performance. Second, with average bandwidth consumption across all memory partitions as metric, the memory intensiveness of an application is determined. Applications are memory intensive if bandwidth consumption is above 20% of our baseline GPU. We propose Runtime Frequency Calibrator (RtFC) mechanism that calibrates the frequency of memory system at runtime after identifying the application type. We evaluate the performance impact of proposed mechanism with baseline GPU and with two techniques of scaling up number of memory partitions and frequency of memory system. All the memory intensive applications experience performance improvement with 66% being the maximum. With RtFC being targeted for memory intensive applications, performance loss is insignificant in most of the compute intensive applications.