how to use all cpu cores to train model in keras

0
votes

I read that keras will automatically use all available cores in my cpu. But my one does not seems to do so. I use google platform's Jupyter notebook. I tried the virtual machine with 8 core, 16 core and 32 core. But training speed is same(or at least I can not see any difference). I wonder how can I use all cpu cores to train the model?

1
tensorflowkeras
Your results do not mean that multiple cores are not being used, it means that computation is not the bottleneck. - Dr. Snoopy

1 Answers

1
votes
  • GPUs has 10^3 cores. That's 100 times more than a CPU has.
  • The bandwith between those cores is much larger and faster compared to a CPU
  • They also have faster and larger memory to store data and intermediate results.

A x2, x4 or even x10 upgrade on your CPU will not yield significative improvements.