Amazon AMI refers to an Amazon Machine Image (AMI) template that consists of specific software configurations including an operating system. An AMI template can be used to launch a machine image instance using Amazon Elastic Computing Cloud (EC2) resources. The Amazon Web Services (AWS) management console already provides a set of AMI templates specifically configured with recent deep learning tools in order to quickly enable learning from big data. As shown in the example below those AMIs are usually named with a specific version number using also a specific operating system. Users of the AWS EC2 service can launch an instance of such an Image in a couple of minutes just specifying the concrete compute resource and storage resource to be used in conjunction with this image instance. Users of those below shown deep learning AMI templates should choose GPU computing resources since they enable faster learning from data.
As the image above shows there are a wide variety of known deep learning tools pre-installed and pre-configured. Examples include MXNet, TensorFlow Tool, PyTorch, Keras, Chainer, Caffe/2, Theano and CNTK. This in turn saves much time of users that have to install it since the installation of a deep learning framework is in many cases not straightforward. Reasons for a relatively hard installation of those packages include dependencies on the correct lower Level library versions like cuDNN or support for specific types of GPUs. Hence a key benefit of using AMI templates are that the above mentioned deep learning frameworks are already configured with NVIDIA CUDA, cuDNN, NCCL and Intel MKL-DNN. Those later lower level libraries enable a more efficient use of deep learning frameworks. Please refer to our article on deep learning AMI for more pieces of information of how those AMI instances are used.
Amazon AMI Details
Please refer to the following video for more details on this topic: