Performance of Single-processor Blas on Ibm P690 Performance of Single-processor Blas on Ibm P690

The Basic Linear Algebra Subprograms, BLAS, are the basic computational kernels in most applications. BLAS 1 and BLAS 2, the vector-vector and matrix-vector routines, require memory accesses in the same order as computations and thus cannot achieve performance close to peak performance on modern computer architectures. BLAS 3 matrix-matrix operations on n× n-matrices on the other side can do order n operations with only order n memory accesses. This much better ratio of computation to memory access allows for much higher performance. To show which performance can be expected using the BLAS routines from IBM’s ESSL on an IBM p690 we investigated the performance of one routine of each BLAS level and compared it to that of the corresponding routines on a CRAY T3E.