I currently have a problem with the speed of training a new tensorflow model. Actually, I assumed that if I use more than one GPU for training, the speed of training will in
