Using GPUs in a notebook (Azure)

This article outlines quota requirements for provisioning GPU VMs, error messages you may encounter due to quota limits, and steps you can take to increase your quotas.

Terra on Microsoft Azure is pleased to offer the computational power of NCSv3 (NC6s_v3 and NC12s_v3) series GPU VMs. The ability to use GPUs makes Terra the ideal platform to run a range of biomedical analysis applications such as Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL). See more details in this blog post or the Use GPU VMs on Terra on Microsoft Azure featured workspace. 

Overview: Subscription GPU quota requirement

To successfully launch a GPU VM, the workspace’s Azure subscription must have a quota allocation for GPUs.

Step 1: Request a quota increase

To use GPUs in Terra, Azure subscription owners will first need to request a GPU quota increaseThis is because Azure subscriptions start with a GPU quota of zero by default. Note that only subscription owners can request a quota increase.

Step-by-step instructions

The Azure subscription owner can request a quota increase by following the directions in this Microsoft Azure article.

Step-by-step summary

1. Sign in to the Azure Portal.
2. Select Subscriptions under the Navigate menu.
3. Select the desired subscription.
4. Select Usage + Quotas in the left-hand menu.
5. Search for and select the desired VM type (i.e., NCSv3)
6. Click on Request Quota Increase.

How much should you request?We recommend you request 12 cores to start. Requesting too much quota may trigger an investigation.

Step 2: Provision GPUs in an Azure Virtual Machine

2.1. Go to the workspace Analyses tab.

2.2. Select the appropriate Cloud Compute Profile by clicking on the cloud icon in the right side panel and Settings under the Jupyter section.

2.3. In the Azure Cloud Environment configuration pane, click on the Cloud Compute Profile drop-down.

2.4. Select either the Standard_NC6s_v3 or Standard_NC12s_v3 configurations (Image below).

2.5. Click the Create button at the bottom right-hand side.

Troubleshooting quota errors

If your GPU-powered VM fails to create, you will get an “Error loading cloud environment” message (see screenshot below).

What to do (step-by-step instructions)

1. Check with the subscription owner to make sure you have a GPU quota.

   Note that all MS subscriptions start with zero GPU quota.
   
   If you don’t have a GPU quota, Azure subscription owners (only) can request one by following the instructions here.
   
   

2. Try a different GPU type, or wait a little and try again.

   If the GPU quota has been temporarily met, you can choose to wait for a GPU to be released. The demand for GPUs will increase or decrease throughout the day, and will also depend on the specific Azure region.
   
   The best bet is to try a different GPU type or to just wait and try again.
   
   

3. Request a quota increase

   If trying a different GPU type or waiting doesn’t resolve your quota issues, you may need to request a quota increase. The Azure subscription owner (only) can request a quota increase by following the directions in this Microsoft Azure article.