The performance chart of a novel, parallel algorithm designed to find the most important and central nodes in a social network on a single Xeon Phi accelerator. a GTEPS: the number of social-network friendships per second (in billions).
High Performance Computing studies on obtaining the best utilization of the hardware at hand for large and time-critical computing tasks. The continuous architectural evolution of the computer clusters, network infrastructures, multi- and many-core processors and accelerators necessitates heterogeneous computation and novel parallelization techniques and optimized, topology-aware implementations of data analytics kernels to minimize the energy usage and computing costs and to obtain the best performance.
