ИСТИНА

The main goal of this work is to evaluate the performance differences of various data-intensive applications for IBM Minsky and Newell systems. As an important example of data-intensive applications, several fundamental graph algorithms have been selected. According to already existing implementation approaches, efficient implementations of selected graph algorithms have been developed. The measured perfor- mance has been compared between two generations of IBM Power sys- tems, demonstrating the 1.5-4 times better performance on the Newell system. In addition roofline models have been generated for both Power and NVidia implementations, which allowed to compare the efficiency of the developed implementations for both systems. Based on the results demonstrated in the paper we can conclude that with a well-optimised code for the Minsky system, it is relatively easy to obtain significant acceleration with high efficiency on the Newell system at least for the described class of graph algorithms.