From social networks to protein networks to financial transactions, graphs are everywhere. Graph Analytics represent a key tool for data science to take advance of this type of network information. Many “Bigdata” and NoSQL techniques for analysis and data science that work well for relational and structured data, do not scale effectively when applied to challenges in graph analytics and traversal algorithms. The data locality and graph access patterns challenge existing HW architectures and place a premium on bandwidth to main memory.GPUs currently have 10X advantage over CPUs in this area.
The advantage is projected to grow to 100X by 2016. This talk will discuss why GPUs are game-changer by dramatically improving the price-performance ratio for very large graph analytics over existing technologies. It will present results for work in GPU Acceleration of graph analytics within both research and industry applications.