Computational Cluster Programs

BLAS BENCHMARK -- SGEMM and DGEMM

Overview

SGEMM and DGEMM compute, in single and double precision, respectively:

C := alpha*op( A )*op( B ) + beta*C
where:
  • A is M-by-N, B is N-by-K and C is M-by-K.
  • alpha and beta are scalars.
  • In the following tests, we use
    op( A ) = A op( B ) = B

All cases were run on a single processor on one of the Hoffman2 Cluster compute nodes. The code is single-threaded and statically linked.

MFLOPS is calculated as:

MFLOPS = (1x10**-6) * 2 * N**2 / (CPU seconds)

Results

Versions of BLAS compared: BLAS library from the Netlib Repository, ATLAS library, Intel-MKL library, AMD ACML Library and Goto BLAS.