<div>Modern supercomputers like Shaheen II at KAUST are specially designed and architected for large-scale parallel simulation workloads such as computational fluid dynamics or weather forecasting.  However, many Big Data Analytics algorithms and workloads can be run very effectively on supercomputers, once the underlying differences in computer architecture and programming models are taken into account.  In this talk we will discuss the fundamental architectural differences between ordinary servers and clusters where many analytics codes are developed, and supercomputers.  We will then discuss how this applies to specific applications, including a large-scale semantic graph (RDF) database and more general graph algorithms.  We will look at parallel analytics packages such as Spark and R, and large-scale Deep Learning training with TensorFlow.  Finally, we will cover the performance that can be obtained through this approach.</div>
