Once you have linked your AWS account with Clouderizer, next step is to start running your projects on AWS instances. Below are the steps to do so
- From Clouderizer console, create a new project or modify an existing project.
- Go to Machine tab and select AWS EC2 instance
- You can specify Instance Type here. 3 GPU and 2 CPU options are listed to choose from. You can also specify the SSD volume size.
- Whenever you select or change the instance type, Clouderizer scans the current bidding price for this configuration across all AWS regions and suggests you cheapest region to spin this instance in.
- Bid Price : Recommended bidding price is 30% more than the current bidding price in the cheapest region. This is to ensure you can work on your project without interruptions due to slight variations in Spot bidding prices. You can always override and specify bid price of your choice as well here.
- Region : Region recommendation can also be overridden. Once you modify the region, we auto calculate the best bidding price for the region.
- AMI : By default Clouderizer uses AWS Deep Learning Ubuntu 16.04 AMI which is bundled with most Nvidia GPU drivers. You can override this with an AMI or your choice. Please note Clouderizer projects will run inside Clouderizer define Docker container comprising of most modern Deep Learning frameworks like Tensorflow, Torch, Theano, Keras, etc.
- Once you have configured your project, press Next and fill up details on Setup and Workspace tab as per your project requirements. Finish the wizard.
- Now to run your project, go to main console and press Start for the project created above.
Clouderizer will now create a new Spot instance and run your project inside it. Once project initialization is complete and status changes to Running, you can start working on it. Remote Terminal, Jupyter Notebook, Tensorboard and Clouderizer Serving buttons will become enabled (as per your remote access settings) and can be used to access your AWS instance.
Stopping the project from Clouderizer console will also terminate the AWS spot instance automatically.