How can i request a GPU cluster on Azure Machine Learning?

Question

Trying to request a GPU cluster, but in the quotas page, there is no GPU's available for me to select.

Answer 1

Hi @Alex

Thank you for reaching out to Microsoft Q&A. I understand you are looking to request a GPU cluster on Azure Machine Learning.

GPU quota in Azure Machine Learning is requested for a specific GPU VM family in a specific Azure region; it is not a general GPU-cluster quota.

From my experience, I would first create a compute resource in Azure Machine Learning Studio and choose Select from all options. This view can show GPU SKUs that currently have zero quota. Select the required SKU and use the displayed quota-request link.

You can also submit the request from the Azure portal:

1. Open the Azure Machine Learning workspace.
2. Go to Support + troubleshooting > Usage + quotas.
3. Select the subscription and filter to the region where you plan to create the cluster.
4. Select the required GPU VM-family SKU, enter the new core limit, and submit the request.

If the GPU family is not listed, I suggest confirming the subscription and region filters, then checking Select from all options in Studio. GPU VM families commonly begin with zero quota, and quota approval does not guarantee available capacity.

Reference: Manage and increase quotas and limits for resources with Azure Machine Learning

Please 'Upvote' (Thumbs-up) and 'Accept' as answer if the response was helpful. This will be helpful to other community members who face the same issue.

Best regards,

Andrew S Taylor

Answer 2

Hello @Alex GPU VM families in Azure Machine Learning typically start with zero quota by default, which is why you may not see any GPU options available when creating a compute cluster. To request GPU quota:

1. In Azure Machine Learning Studio, go to Compute and start creating a compute resource.
2. Switch to GPU sizes and select "Click here to view and request quota".
3. Choose the required GPU VM family (for example, NCasT4_v3, NC_A100_v4, etc.) for your target region and submit a quota increase request. [learn.microsoft.com]

You can also submit the request from the Azure portal under Usage + quotas by selecting the relevant GPU VM family and requesting additional cores. Quotas are managed per VM family and per region.

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Answer 3

For Azure Machine Learning, GPU VM families usually start at zero cores by default, so they may not appear as available until quota is granted for that VM family in the target region.

Use this path to request it:

1. In Azure Machine Learning studio, go to Compute and start creating a compute resource.
2. Switch to GPU sizes.
3. Select Click here to view and request quota.
4. In the quota view, select the GPU VM family needed for the cluster, such as NCasT4_v3 or another GPU family available in the region.
5. Enter the new core limit and submit the request.

An alternative path is:

1. Open the Azure portal.
2. Go to Usage + quotas.
3. View the current quota limits for the region.
4. Select the VM family SKU to request an increase.
5. Enter the new limit and select Submit.

Important details:

• GPU quotas default to zero cores in Azure Machine Learning for specialized GPU VM families such as NCasT4_v3, NC_A100_v4, and NDv2.
• Quota is per VM family and per region, so check the region where the compute cluster will be created.
• If no GPU family is selectable, that usually means there is currently no quota assigned for that GPU VM family in that region.
• If the subscription is Free/Benefit/Sponsorship, quota increases for GPU instances may not be available. The documented Q&A guidance is to upgrade to Pay-As-You-Go.
• Even after quota is approved, GPU creation can still fail if the region has temporary capacity constraints for that GPU SKU.

If the goal is a compute cluster specifically, Azure Machine Learning compute clusters support both CPU and GPU compute nodes for training and batch inference.

---

• Manage and increase quotas and limits for resources with Azure Machine Learning
• Manage and increase quotas and limits for resources with Azure Machine Learning
• Create an Azure Machine Learning compute cluster
• Unable to create Standard_NC4as_T4_v3 in Azure Machine Learning. Getting Standard_E4ds_v4 instead. - Microsoft Q&A
• can't create GPU instance & can't request quota in Machine learning - Microsoft Q&A
• Cannot start VM in azure machine learning workspace while creating a GPU instance - Microsoft Q&A