The difference between OpenBLAS and MKL

1

I want to compile the source code in ARMv8 machine, and MKL does not support ARM.Then, I choose OpenBLAS as the BLAS.However, when I compile it successfully, I found the CPU utilization is low.My machine has 128 cores but when I run the following benchmark:

import timeit
runtimes = []
threads = [1] + [t for t in range(2, 49, 2)]
for t in threads:
    torch.set_num_threads(t)
    r = timeit.timeit(setup = "import torch; x = torch.randn(1024, 1024); y = torch.randn(1024, 1024)", stmt="torch.mm(x, y)", number=100)
    runtimes.append(r)
# ... plotting (threads, runtimes) ...

The CPU untilzation just like this and time has hardly changed.
image

The compile step just like this:
export USE_OPENMP=1
export BLAS=OpenBLAS
export USE_CUDA=0
python setup.py install

machine:aarch64

Can OpenBLAS do parallel computing on Pytorch?If not, how can I do to improve the performance?

1 Like