Get Started with Machine & Deep Learning Models & Frameworks Quickly with Oracle GPU Instances

In this post, we will learn how quickly we can set up Oracle Cloud Infrastructure GPU Instances, Get access to NVIDIA GPU Cloud - Container registery and spin up TensorFlow and various other Machine & Deep Learning Frameworks.

Oracle Cloud Infrastructure GPU Instances uses NVIDIA Machine Images. These images have preconfigured NVIDIA GPU CLOUD (NGC) set up. What this means is, we can have access to NGC Container Registry and run any of the Framework Containers & Models availble on NGC  -  https://ngc.nvidia.com/catalog/containers

Lets Get Started.

First we would need to Spin Up GPU Instance in Oracle Cloud Account.

Let us jump straight to OCI Console and Create a GPU Instance. We gave it a name AlphaGPUDemo

Next step is to Select a Shape. For GPU Instances, we will select NVIDIA GPU Cloud Machine Image from Oracle Images. This image is Ubuntu based.

Next Step is to select Shape. Can select any GPU Shape from your options. Here we have selected VM.GPU3.1 here.

Add the public key for the instance.

Click Create and new Compute GPU instance with NVIDIA Image will be created and started. Just a note, do add the instance to Internet Facing VCN. If there is no VCN, there is option to create a default VCN with Internet access when you create GPU Instance in above screen.

Ok, So instance is all set, up and running.

Now next step is to create account at NVIDIA NGC. Log on to https://ngc.nvidia.com/ and create a new account.

Once the account is ready, login and Click on Set up on the left navigation bar.

Select Generate API Key.

Make a note of new generated API Key. This will be used to authenticate against NGC from new Compute GPU Instance. For security reasons, this is being masked.

Now grab public IP of GPU Compute Instance and login using ssh. Use the username as Ubuntu. It asks for NGC API Key. Paste the key which was generated in the previous step.

Login will be successful.

Now go to https://ngc.nvidia.com/catalog/containers and select the framework you need to use for running your Machnie Learning Models. There are large number of popular frameworks in NGC Registry all availble to run as Container.

We will download and run the popular TensorFlow for this example. Select TensorFlow and copy the pull command and run it in the shell.

The command will download TensorFlow Container Image from NGC registry. 

Now run the container using below command. Container will get started and we will get access to it, since we are running container in foreground.

Now we can use TensorFlow for any of our analysis.

We will run the sample MNIST example. Go to the below path.

And run # python mnist_with_summaries.py

And here we go, TensorFlow in Action.