speed up training with multiple cores - possible?

jcjohnson / fast-neural-style

Feedforward style transfer

4.28k stars 814 forks source link

speed up training with multiple cores - possible? #138

Open monkey-jsun opened 6 years ago

monkey-jsun commented 6 years ago

I'm currently running a training session on my own server which does not have GPU. I pretty followed the README file exception with my own style image. It runs pretty slowly. So far only 40 iterations have completed for about 12 hours of running.

I notice that training seems to only use 1 CPU core while my server has tons of CPU cores. Is there a way to speed up the training with multiple cores?

jcjohnson commented 6 years ago

Check the BLAS installation that your torch7 install is using:

https://github.com/jcjohnson/neural-style/issues/88

arunasank commented 4 years ago

@jcjohnson possible to engage multiple GPU cores?