How do I apply mini-batch training with the normal neural network training functions?

Hi, I am using the scaled conjugate gradient training function for a classification problem with a very large number of samples (15*10^6). Because of the large training set, I am running into memory errors on my GPU. Please, how can I apply mini-batch training in this instance, to circumvent the problem? I have been digging and digging, but could not find an answer.

Peter