Performance rating via the FEAST INDICES

Based on algorithmic and computational studies, we present the FEAST INDICE S which are new indicators for the realistic performance of many recent pro cessors, as typically used in 'low cost' PC's/workstations up to (parallel) supercomputers. Since these tests are very specifically designed for the e valuation of modern numerical simulation techniques, with a special emphasi s on 'large scale' FEM computations and iterative solvers of Krylov-space/m ultigrid type, they examine new aspects in comparison to the standard bench mark tests (LINPACK, SPEC FP95, NAS, STREAM) and allow new qualitative and particularly quantitative ratings of the various hardware platforms. We exp lore the computational efficiency of certain Linear Algebra components, her eby applying the typical sparse approaches and additionally the sparse band ed style from FEAST which enables us to exploit a significant percentage of the available Peak performance. The tests include 'simple' matrix-vector a pplications as well as complete multigrid solvers with very robust smoother s which are necessary for the efficient treatment of highly adapted meshes.