|Name:||(13) Splotch on the Xeon Phi|
|Time:||Monday, June 23, 2014
05:46 pm - 05:53 pm
CCL - Congress Center Leipzig
|Speaker:||Tim Dykes, University of Portsmouth|
|Abstract:||The management and analysis of modern large-scale datasets generated by scientific experiments (observations or numerical simulations) can be very challenging due to continuously increasing sizes and complexity. Traditional data mining and analysis methods can be expensive if applied to large-scale datasets.
Visual exploration and discovery can represent invaluable tools providing scientists with prompt and intuitive insights, enabling them to identify characteristics and manageable regions of interest for applying time consuming methods. They can also be effective means for communicating scientific results not only to researchers but also to members of the general public.
Splotch is a rendering algorithm for large-scale datasets exploiting modern High Performance Computing (HPC) architectures. Parallelism on HPC systems is achieved by deploying the MPI, OpenMP and CUDA paradigms. In order to address emerging technologies and exploit all potential hardware within a HPC environment the algorithm is extended, building upon the MPI and OpenMP implementations, to create an offloading model exploiting Intel’s Xeon Phi using the new Many Integrated Core (MIC) architecture and the provided C++ offload extensions.
Tim Dykes, University of Portsmouth; Claudio Gheller, ETHZ; Marzia Rivi, University of Oxford; Mel Krokos, University of Portmouth; Klaus Dolag, University Observatory Munich; Martin Reinecke, Max Planck Institute for Astrophysics