Graphs are not just nodes or links. They are powerful data structures anyone can use to represent complex dependencies in their data. Graph applications are used in various places ranging from cancer research to large-scale cyber threat detection to collaborative filtering recommendation systems. In the world of data-intensive analytics, memory bandwidth is the primary performance restrictor. Because graph algorithms display non-locality and data-dependent parallelism. When you crisscross a large group, you are constantly asking for from main memory. For these problems, GPUs provides superior bandwidth to memory and can deliver significant speedups over CPUs. GPUs are very fast for graph processing and analytics, where memory bandwidth is a problem. The memory bandwidth of GPUs provides a new way to speed up data-intensive analytics and graph analytics. For more read the article written by Brad Bebee ( CEO, Blazegraph) : https://devblogs.nvidia.com/parallelforall/gpus-graph-predictive-analytics/