The spread of deep learning methodologies has coincided with the popularity of the GPU in computing. These methods are very amenable to parallelization. Indeed, a lot of deep learning methods would not be feasible without GPUs.
Running deep learning models on the GPU requires the installation of the CuDNN library from NVIDIA. This library contains fast implementations of the low-level mathematical primitives needed for deep learning systems. You'll need to register with the NVIDIA Developer Network and then download the library from https://developer.nvidia.com/cudnn. Choose the version for your operating system and install it on your machine. You'll need to have installed the appropriate graphics drivers and the CUDA toolkit prior to installing CuDNN.
These primitives are now ready to be used from a higher-level library. To demonstrate...