Performance of a Supernodal General Sparse Solver on the CRAY Y-MP: 1.68 GFLOPS with Autotasking

摘要

Numerical results with a general sparse linear equation solver package are presented. This software has been developed for production use by researchers at Boeing Computer Services and is available from Cray Research for Cray X-MP systems. Recently this package has been ported to the new eight processor Cray Y-MP. Initial results using microtasking and autotasking primitives and all eight processors yielded 1682 MFLOPS for the factorization of a sparse matrix of order 24,565. The results reported here demonstrate the high performance levels obtainable on the new Cray Y-MP in autotasking mode. We also believe that this is the highest speed ever obtained factoring a non-trivial sparse matrix.