BLAS performance is very system dependent, so it’s best to run the tests on the machine you want to use. Since there are only a few implementations of BLAS, this works less than it seems (usually an implementation of hardware vendors , ATLAS and GOTO BLAS ).
But note that BLAS only covers dense matrices, so for sparse matrix multiplication you will need Sparse-BLAS or some other code. Here, the performance will differ not only depending on the hardware, but also on the sparse format that you want to use, and even on the type of matrix you are working with (things like sparseness, bandwidth, etc.). Therefore, even more than in the dense case, if you need maximum performance, you will need your own tests.
jdisk source
share