Introduction

The NVIDIA compute software “stack” consists of various software products in the system software or infrastructure that are required to bootstrap a system with NVIDIA GPUs and be able to run accelerated AI or HPC workloads. A software architecture diagram of CUDA and associated components is shown below for reference:

Figure 1 Overview of CUDA Toolkit and Associated Products

While NVIDIA provides a very rich software platform including SDKs, frameworks and applications, the focus of this document is on drivers, CUDA Toolkit, and the Deep Learning libraries.