Getting Started with Google Cloud Platform for solving Deep Learning Problem

I went through the hectic job of searching all the internet to find the right solution for creating a simple Virtual Machine to run my Python program.

I’m here to guide anyone interested to use the plethora of services provided by the Google Cloud Platform(I can’t believe that it was released in 2008)

Anyways to let’s get started,

First you will need a Gmail Account.

Then Click on this Link :

In there, click on TRY FREE

Log in to your G-Mail Account.

Then is a very tricky part where you’ve to setup your payment account. Use your Credit Card or Debit Card to register for it. Sometimes, they charge a meager fees 0.0001 dollars(or 2 ruppees) as just to verify your account. After verification is the time to see some money. Let’s see now, 300 dollars(Rs.21,988/-) are added to your account as free credits !!! (Smug grin :-P )

The fun stuff begins now

Setting up the VM Instance:

Baby Step 1

Next, you’ll be directed to this page where you’ll click on CREATE :


Here you can configure your Virtual Machine with any software or hardware you like. Say for deep learning a small model,Dense Net on Cifar 10 database,we can use a 2 core CPU with 8GB Memory and 1 Tesla P100 GPU. This configuration really helped me improve the speed of training in comparison with Google Colab. You can choose whichever configuration you like. Remember you shall be billed accordingly.(Initially upto 300 dollars its free,Google will not bill your card)

Now you get to choose your OS. For deep learning practices I would recommend choosing the Intel Optimized Deep Learning Image: TensorFlow with Debian Linux Image. This image replicates all the basic software thats required for creating a deep learning model like python,jupyter,pip etc. I like the fact that immediately after the install it asks us to install the NVIDIA CUDA driver for the chosen GPU. And then be sure to choose the size of the boot disk and as an SSD(gives a really small cost differance with the standard disk of 0.001 dollars).

Click on allow HTTP traffic and HTTPS traffic . And Create! There you’ve your instance up and running.

Click on SSH to get started!

You’ll notice a screen like this,

The first thing you’d want to do is, hit y for installing the nvidia driver.


Execute the following commands

sudo apt-get update

Now let’s configure the network on the cloud to allow us to run jupyter notebook.

Run the following commands

jupyter notebook –generate-config

sudo nano ~/.jupyter/

It’ll open this

c = get_config()
c.NotebookApp.ip = ''
c.NotebookApp.open_browser = False
c.NotebookApp.port = 4808

Write this code as shown in the red lines window.

Then you have to press

Ctrl + O

Enter key

Ctrl + X

Now create a password for your jupyter notebook by command

jupyter notebook password

Then minimize the SSH terminal window, click on instance

Click on default in Network Interfaces

Then click on firewall rules and add firewell rule

Name your firewall rule and specify target as All instances in the network

Add source IP range as

and TCP protocol as anything you like (eg.4808)

Back too the SSH terminal type this code

jupyter-notebook --no-browser --port=4808

Instead of 4808 write your own TCP protocol number.

Then go to the Chrome page and copy the external IP the one with ephemeral in brackets.

Goto a new chrome tab and paste your external IP address. then give a colon and type in your port number.


(x,y,z are integers)

Then you’ll see this screen : and you are good to go :-D

Source: Deep Learning on Medium