It’s very difficult to implement algorithms optimally for GPUs. Therefore NVIDIA provides the library cuDNN for deep learning with neural networks. This article published in the german magazine iX Developer gives an overview of cuDNN, shows an implementation of its usage and explains advantages and disadvantages.
The NVIDIA CUDA Deep Neural Network library (cuDNN)
is a GPU-accelerated library for deep neural networks. It is used by many of the well known machine learning frameworks, e.g.
TensorFlow, Caffe2, MicrosoftCNTK, and pyTorch.
The article is written in german and is called “Unter der Haube - cuDNN schafft die Grundlage für Deep Learning mit GPUs” which means “under the hood - cuDNN as the basis for deep learning with GPUs”.