We can easy to set up a training environment from a notebook with "click" for elastic of CPUs/GPUs * Connectivity and easy to deploy – AWS SageMaker is AWS managed service and it easy to integrate with other AWS services inside of a private network. Which also impact to big data solution, ETL processed with data can be processing inside of a private network and reduce cost for the transfer. AWS managed service will help to reduce the resource we need to create.

"Build" phase in sage maker allows to launch instance with a JupyterHub for creating a notebook. AWS IAM role needs to be created to have access to the data in S3 from sage maker notebook. Notebook instances are available in three variants -- t2.medium, m4.xlarge and p2.xlarge. In my view, we should be able to launch any type of instance based on our choice or work load. May be future versions of sage maker might support this..:) Each notebook instance is installed with Anaconda Python packages along with Tensorflow and mxnet.