Abstract: The applications which are running on heterogeneous computing system (HCS) with hybrid processors (CPU and GPU) often uses only one processor and the Other processor will be in ideal state for the rest of operations of the application and this results in the wastage of the available computational resources and issues related to performances. It is possible to avoid this kind of wastage of the computational resources of modern HCS by dividing work across hybrid processors of HCS. We introduce a technique to fully utilize the hybrid processors of HCS to provide the significant improvement in performance and usage of hybrid processors for matrix multiplication based applications. We are using library functions like cblas (MKL) and cublas (NVidia’s CUBLAS Library function) to divide the work across the hybrid processors of HCS.
Keywords: scale, GPGPU, hybrid processors, HCS