Pylearn2 out of memory error

I am trying to train NN using pylearn2
I use stochastic gradient decent.
I had always a message about failed GPU memory  allocation in
train->self.sgd_update(*batch)
To fix it, I reduced size of the batch in yaml-config:
batch_size: 200
was replaced by
batch_size: 100