Poor openBLAS performance for large matrix multiply?

New to Julia
3

Set the environment variable OPENBLAS_VERBOSE=2 before starting julia, to see what coretype is OpenBLAS using. Alternatively (but it’s more verbose):

julia> using LinearAlgebra, OpenBLAS_jll

julia> strip(unsafe_string(ccall((BLAS.@blasfunc(openblas_get_config), libopenblas), Ptr{UInt8}, () )))
"OpenBLAS 0.3.23  USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell MAX_THREADS=512"

Also, check with LinearAlgebra.BLAS.get_num_threads() the number of threads when you’re running the code with OpenBLAS vs MKL, in case there’s any difference, and if so you can change the number of threads with LinearAlgebra.BLAS.set_num_threads.

4 Likes