We are implementing a simple yet real-world deep neural network having 3 layers, having one layer for input one layer for output and one hidden layer that can be trained on various data sets and will produce results. We are planning to implement neural network by making use of libraries and frameworks like Theano, Tensor Flow, PyCuda, Hebel, Chainer etc. These libraries allow C/C++ and Python code to be implemented directly on GPUs. We plan to train and test this neural network on datasets like MINST (contains large dataset of images of handwritten digits), Iris Flower Dataset (contains images of ﬂowers), CIFAR (contains 60000+ small images of various objects), SVHN (contains real world street view images) etc. We will be training the neural network on the above mentioned datasets using varying precision (8 bit, 16 bit, 32 bit, 64 bit).