Closed snayfach closed 3 years ago
This was really confusing to me as well, but disabling multiprocessing completely still used all my cpus. Turns out the theano backend to keras automatically uses all available cores for the internal matrix operations.
For a fix: http://deeplearning.net/software/theano/tutorial/multi_cores.html
or just export OMP_NUM_THREADS=4
to get max 4 threaded operation
I've noticed that by default DVF uses multiple cores (up to 20 on my machine) and that the flag
-c
or--core
does not change this behavior. Do you know why this might be happening?