A pre-configured and fully integrated software stack with TensorFlow, an open source software library for machine learning, and the Python programming language. It provides a stable and tested execution environment for training, inference, or running as an API service. The stack can be easily integrated into continuous integration and deployment workflows. It is designed for short and long-running high-performance tasks and optimized for running on NVidia GPU. The stack includes CUDA, a parallel computing platform and API model; and cuDNN, a GPU-accelerated library of primitives for deep neural networks. It also includes NVidia drivers and Development preset, program development and building tools, including C compiler, make etc.

Versions: TensorFlow 1.10.0, Python 2.7.14, CUDA 9.2.148.1, cuDNN 7.2.1, NVidia drivers 396.51, Development preset 1, Libc 2.22, OpenBLAS 0.2.20, Python_enum34 1.1.6, NumPy 1.13.3.