How to use all available cores in mxnet


#1

Hi, have the following question:
I’m gaining first experiences with neural networks.
I started with keras and tensorflow but then I read, that mxnet is much faster than tendorflow. Therefore I installed keras_mxnet-2.2.2 and mxnet-1.3.0.
I changed the keras-backend to mxnet and I tried to train an CNN.
But it was much slower than with tensorflow, because only one of my 8 cores was used, while tensorflow uses all 8 cores.
What can I do to use all available cores?


#2

uninstall mxnet using pip uninstall mxnet -y and install the MKLDNN version that uses all the cores on CPU, by using pip install mxnet-mkl.

Though the performance boost on CPU vs tensorflow is smaller than the one you could see on multi-gpu.