SUB: NEED HELP TO SETUP THEANO/KERAS TO USE ALL CORES OF CPU.
I intend to use my local windows machine for working on small sample datasets and use the AWS GPU server for final runs in order to keep my AWS costs low.
Finally, I have successfully run Lesson 1 notebook on both Windows and AWS after fixing the initial few errors. On a 1000+ images dataset, my windows laptop executes the VGG.fit() statement in 3 hours Vs the AWS GPU doing it on all 22000 images in about 10 minutes!
My problem is that although the windows machine is a multicore CPU, the python/theano/keras uses only one core. I did set the environment variable OMP_NUM_THREADS=7, hoping that it will make all 7 cores get utilized.
Is there anyway to decrease the time taken on a CPU by making it run parallel on all cores? I understand it will still be 10X the time taken on a GPU, but should be useful for trying code on a sample dataset.
Any suggestions would be appreciated