The FEAST indices - Realistic evaluation of modern software components andprocessor technologies

We examine the computational efficiency of linear algebra components in ite rative solvers for grid-oriented simulations of PDEs. While tile standard s parse matrix-vector (MV) techniques show significant losses of performance, especially on modern processors. our sparse banded components have the pot ential to exploit today's high computing power. We explain the major concep ts of the FEAST software which contains such highly tuned numerical linear algebra basic components (Sparse Banded Bias) up to complete multigrid solv ers, all being optimized with respect to the actual hardware platform. Base d oil algorithmic and computational studies, we present the FEAST indices w hich are indicators for tile true performance of many modern processors, de pending on tile underlying FEM space. tile problem size and the implementat ion style, These indices allow a new rating of the various hardware platfor ms with regard to different mathematical solution strategies, for academic and realistic numerical problems and ranging from 'low cost' PC's up to sup ercomputers. (C) 2001 Elsevier Science Ltd. All rights reserved.