AWS Deep Learning AMI are built and optimized for building, training, debugging, and serving deep learning models in EC2 with popular frameworks such as TensorFlow, MXNet, PyTorch, and more. Deep learning frameworks are installed in Conda environments to provide a reliable and isolated environment for practitioners. The AWS Deep Learning AMI is provided at no additional charge to Amazon EC2 users. Deep Learning frameworks are pre-configured with latest versions of NVIDIA CUDA, cuDNN and Intel acceleration libraries such as MKL-DNN for high performance across CPU and GPU AWS EC2 instance types.

Below are the core components of AWS Deep Learning AMI:

• Popular deep learning frameworks includng TensorFlow(1.x, 2.x), PyTorch(1.x), and MXNet(1.x) performance tuned for using in AWS Instrasturctures. They are organized into Conda environments that are configured to be used out-of-the-box.
• Built-in support for AWS Elastic Inference.
• Built-in support for AWS Inferentia with the Inf1 instance family.
• AWS Deep Learning Tools including AWS Elastic Fabric Adapter(EFA) and AWS Neuron.
• NVIDIA Deep Learning Softwares Including NVIDIA GPU Driver, CUDA Toolkit, cuDNN, NCCL, and Fabric Manager.
• Containerization platforms including Docker, and NVIDIA-Docker for build and run GPU accelerated Docker containers.
• Intel Architecture performance library Intel MKL-DNN.
• A collection of popular tools such as awscli, boto3, numpy, scikit-learn, opencv, pandas, matplotlib, graphviz, jupyter, ipython, and more.