KONWIHR

COMPETENCE NETWORK FOR TECHNICAL, SCIENTIFIC HIGH PERFORMANCE COMPUTING IN BAVARIA

OMI4papps

Optimization, Modeling and Implementation for highly parallel applications

In the emerging era of multi-/many-core technologies an improvement of computational speed for a specific application or numerical problem can be gained either by new numerical methods or by code optimization and parallelization. The latter approach often requires deep insights into computer architectures, optimization strategies and parallel programming models as well as parallel libraries. Most numerical research groups can not keep track of the rapid changes now happening with the advent of multi-/many-core technologies and require comprehensive highlevel support to address the problems arising from massive parallelism, shared onchip resources or ccNUMA data locality, to name only a few. This KONWIHR-II project addresses those problems and is intended to provide a central HPC user support resource for HLRB and KONWIHR projects and other research groups that rely on computationally intensive simulations. The project is hosted by the HPC support groups of the Computing Center Erlangen (RRZE) and Leibniz Supercomputing Centre (LRZ). These support groups have proven special expertise in parallelization and optimization of user applications, programming and evaluating new (highly parallel) architectures, optimizing data access as well as establishing performance models for complete applications and numerical kernels.