Create an AWS Account First, get your AWS ID

Visit and select Student
https://www.awseducate.com/Application

Don't create a Starter Account! Instead enter your AWS ID for the account that you created earlier. 

Once you receive your email with Promo Code, add it via "Credits" in the Billing Dashboard

Go to the main console page, and change region to US East Nothern Virginia. 

Go to Support (top right of menu bar), and create a new support case. Request a Service Limit Increase for EC2 Instance, in Northern Virginia. Then select p2.xlarge. You can select other regions, but approval may take much longer. 

You may have to wait a few hours until your account is internally confirmed to finish the next few steps.

Click Launch a Virtual Machine in the AWS Main Console (make sure you're in N. Virginia Region!). Select Deep Learning AMI (Ubuntu) Version 17.0 (or something related) for your AMI, just a few rows down. 

Select p2.xlarge for your Instance Type. (the GPU in g2.2xlarge isn't well-supported by PyTorch anymore).

For Instance Details, you can leave most of the details as-is. One way to extend your effective credits is to Request a Spot Instance - your VM can be run for cheaper, but at the cost of intermittent shutdowns.

Your Storage, you only need one SSD. 75GB is the minimum required for g2.2x. You'll probably want >100GB to fit in all your training data later. You pay marginally more for having more storage. 

For Security Group (we've skipped tags), you might want to enter an extra layer of security. If you always work on campus, you can use 18.0.0.0/8 as your ID mask, so only 18.* IPs can SSH into your VM. 

Right before you launch your VM, create a key pair and store it on your disk. Keep good track of it!

Once your instance has been created, you can SSH into it (click the Connect Button to see how). 

To activate Pytorch, run this command: 
source activate pytorch_p36

MAKE SURE TO STOP RUNNING THE INSTANCE ONCE YOU'VE FINISHED RUNNING ON IT
p2.xlarge costs ~90 cents/hour, the costs add up quickly!

Once you're completely finished with the project, terminate it. There's a small long-term cost in keeping data in AWS. 


To copy files onto Amazon VM, see:
https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/AccessingInstancesLinux.html

Copying a large number of files over network may take a very long time. Try zipping a folder before sending for much easier copy.