Performance of the hybrid parallelized ELPA Eigensolvers on BlueGene/Q and Intel Nehalem Cluster JUROPA

For new exascale computers the degree of parallelismwill increase leading to architectures with more and morecores per node and multithreading and SIMD on each core. Dueto the limited memory per node the pure MPIparallelizing strategy may be no longer suited for these newarchitectures.For applications that already use a hybrid parallelization due tomemory limits, a pure MPI implementation will waste several cores forthe eigensolver unless a hybrid version can be used.Newly developed libraries for dense linear algebra computation such asthe ELPA library for the computation of a large part of theeigenspectrum of dense real symmetric and complex hermitean matricesthus offer versions with hybrid MPI and OpenMP parallelization.On Intel architectures it has been shown that this hybridparallelization can perform better than the pure MPI parallel version. OnBlueGene/Q the hybrid version is still experimental, but it showspromising results although the pure MPI version using all cores still performs better.In this talk we will present performance evaluation results of the two eigensolvers of the libraryELPA on BlueGene/Q and compare it to the results on the Intel Nehalemcluster JUROPA