Getting Started with GPU instances in Oracle Cloud

If you are working with artificial intelligence (machine learning or deep learning) projects at some point, you will need to change your CPU instances to GPU instances speed up the training of your models.

Nowadays, most cloud providers offer GPU as a service and you can use it to speed up your projects. Oracle Cloud offers the two most advanced GPU models for your choice: NVIDIA V100 and NVIDIA P100.

In this blog post, you’ll learn how to request and create GPU instances in the Oracle Cloud.

Access your Oracle Cloud account and go to Menu > Administration > Tenancy Details.

Click Service Limits and expand the “Compute” section. In this section, you can see the limits of your account. By default, the limits for GPU instances are 0 (zero).

To increase the limits, you need to go to My Oracle Support and open a service request.

Click Create Service Request.

Select the following from the displayed menus:

  • Service Type: Select Oracle Cloud Infrastructure.
  • Service Name: Select your Oracle Cloud Account.
  • Problem Type: Select Account Provisioning, Billing and Termination, and then select Limit increase.
  • Description: Enter a good description (For example: I want to use VM.GPU3.1 and VM.GPU3.2 shapes for my Deep Learning projects. I will use these shapes to train neural networks that I created using Keras and TensorFlow. My projects spend 3 hours with CPU instances and I want to accelerate it using GPU instances).
  • 1) What is your Tenancy OCID?: Enter your Tenancy OCID.
  • 14) Which Region and Availability Domain are you requesting this for?: Enter us-ashburn or uk-london.
  • 15) What service/resource requires a limit increase?: Select Compute.
  • 16) What is the total limit being requested for the above service?:
    Enter the new total limit (For example: 1 for VM.GPU3.1 and 1 for VM.GPU3.2).
  • 17) What are you trying to accomplish with this increase?: Enter the
    reason for the request (For example: I want to use GPU shapes for my Deep Learning projects).

My service request with Severity = 2 was solved in 5 hours.

Now you can create your GPU instance!

Select the following from the displayed menus:

  • Operating system or image source: Canonical Ubuntu 16.04.
  • Instance shape: VM.GPU3.1.

Click “Show Advanced Options”, and then click Image tab.
Select Gen2-GPU-2018.12.10-0 as your image build.

Click Create!

When the status of instance change to RUNNING, get the Public IP Address and access it.

To validate the instance, run:

nvidia-smi

Now you can use Oracle Cloud to accelerate your AI projects.

Have a good time!

Author: Waslley Souza

Consultor Oracle com foco em tecnologias Oracle Fusion Middleware e SOA. Certificado Oracle WebCenter Portal, Oracle ADF e Java.

2 thoughts on “Getting Started with GPU instances in Oracle Cloud”

Comments are closed.