What is Digits?
Nvidia DIGITS (Deep Learning GPU Training System) is a toolbox combines populer deep learning frameworks like Caffe, Theano and torch, and they are using the GPUs to speed up the training process. Digits makes it easy for data scientists and researchers to experiment different neural network architectures and visualise the network behaviour and performance, and the utilisation of the GPU in real-time. If you have multiple GPUs, you can train more than one network in parallel, or you can use multiple GPUs to train one network. Digits works through a web interface where you can load the dataset, choose the network, visualise the network architecture, choose the GPU(s) and start the training. During the training, you can visualise the accuracy of the model, statistics about GPU(s) utilisation and temperature, and you can start predicting new samples during the training. Finally, when the training is done, you can download the model and use it for new samples.