Cannot use GPU VMs (NC4as_T4_v3) in Azure ML despite having quota - VMs grayed out

Question

Cannot use GPU VMs (NC4as_T4_v3) in Azure ML despite having quota - VMs grayed out

Max B 0

Hello Azure Community,

I'm trying to deploy an Azure ML real-time endpoint with GPU compute but unable to select GPU VMs despite having quota allocated.

Subscription Details

Type: Pay-As-You-Go
Status: Active
Region: West Europe

Quota Status

I have verified quota allocation in Azure Portal → Quotas:

Standard NCASv3_T4 Family - West Europe: 0 of 12 cores available
Standard NCASv3_T4 Family - West US: 0 of 12 cores available

Problem Description

When creating a Managed Online Endpoint in Azure ML Studio:

Navigate to: ML Studio → Endpoints → Real-time endpoints → Create
Register model and environment successfully
On Compute configuration page, attempt to select GPU VM
All GPU VM sizes appear grayed out with message:

"You do not have enough quota for the following VM sizes"
Specifically trying to use: Standard_NC4as_T4_v3 (4 cores, 28GB RAM, NVIDIA T4 GPU)

What I've Verified

✓ Quota page shows 12 cores available for NCASv3_T4 family in both West Europe and West US

✓ Subscription is Pay-As-You-Go

✓ NC4as_T4_v3 is officially listed as available in West Europe region

✓ Model and environment registered successfully

My Question

Is there additional approval or activation needed for first-time GPU VM usage in Azure ML, even with Pay-As-You-Go subscription and available quota?

The quota shows as available but VMs are not selectable in ML Studio. How can I enable access to GPU VMs for ML endpoint deployments?

Any guidance would be greatly appreciated!

Thank you!

2 answers

Your answer

Answer 1

Aryan Parashar 3,380 Microsoft External Staff Moderator

Hi Max B,

I understand how frustrating it can be to have quota allocated but still be unable to select GPU VM sizes when deploying your Azure ML real-time endpoint.

Quota must be requested and approved at the ML workspace level for GPU VM sizes (such as Standard_NC4as_T4_v3) to become available during Managed Online Endpoint creation.

To resolve this, please verify and request the quota directly within the ML workspace:

Navigate to All workspaces → Quotas.
User's image

If no quota is available, request the quota as shown below:

Select the compute family -> Select Request quota User's image

Enter the New cores limit and click Submit
User's image

Please accept this as an answer.
User's image Thank you for reaching out to The Microsoft Q&A Portal.

Max B 0 Reputation points

2025-12-02T13:59:42.1266667+00:00
Thank you for the response! I've reviewed the documentation about zero-default GPU quotas and followed the steps.

I can see GPU VMs in the deployment list with "12 cores available" shown, but they're still not selectable. I've checked my quota allocation:

ML Studio Quota page shows: Standard NCASv3_T4 Family = 16 cores available

Azure Portal Quotas shows: Standard NCASv3_T4 Family vCPUs = 8 cores (today I made a request to support to increase from 0 to 8)

If I understand correctly, I currently have 16 cores available for the first tier quota (general regional vCPU). And 8 cores for the second tier (VM family vCPU). According to the Azure documentation https://learn.microsoft.com/en-us/azure/quotas/per-vm-quota-requests

Now I have no idea what to do next. I don't see any point in increasing the quotas again, since it didn't give any results. Maybe you can advise on something in this situation, or perhaps I missed something.

I am attaching screenshots of key pages:
Max B 0 Reputation points

2025-12-02T16:37:37.1433333+00:00

New update:

It seems the problem is that NC4as_T4_v3 is not available in West europe right now. I got quotas for Standard NCASv3_T4 Family for West US 2, and now I can directly create a virtual machine on NC4as_T4_v3 in West US 2, but I still can't deploy the endoint. The problem is still NC4as_T4_v3 is showing as unavailable due to lack of quotas.

VM interface:

Ml endpoint interface:

PS: I created a separate workspace in Azure ML Studio for West US 2
Aryan Parashar 3,380 Reputation points Microsoft External Staff Moderator

2025-12-03T04:18:47.51+00:00

Hi Max B,
I completely understand how frustrating this situation can be, and I appreciate the time you’ve taken to analyze the issue. You are correct, quota limits are managed at a regional level. The workspace should only be deployed in a region where NC4as_T4_v3 is available to host your endpoint.

To check which compute is available for your workspace:
Select your region (e.g., West Europe) -> Select Configure workspace quota. -> Apply the filter as shown below.

The available quotas ready for deployment will then be displayed, allowing you to confirm which VM families and cores can be used for your endpoint.

Please let me know if you have any further queries.

If my answer helped, please consider accepting it.

Thank you for reaching out to The Microsoft Q&A Portal
Max B 0 Reputation points

2025-12-03T10:34:14.33+00:00

Hi, Aryan Parashar, thanks for the feedback!

I checked which computes are available for my workspace. And I attach a screenshot with the result. It seems that I have access to NC4as_T4_v3, but I still can't create an endpoint with it.

Також я спробував створити compute в ml studio, і тут я чомусь можу обрати NC4as_T4_v3:

So far I'm at a dead end, I don't know what to do next, I think I've tried all the options. I'd be happy to get feedback
Max B 0 Reputation points

2025-12-05T09:44:31.7133333+00:00

Update: I found a solution to the problem (posted it in a separate comment), thanks for the help Aryan Parashar
Aryan Parashar 3,380 Reputation points Microsoft External Staff Moderator

2025-12-05T09:45:53.7066667+00:00

Hi Max B,

Since the initial deployment issue has now been resolved, could you please create a new thread for the endpoint deployment issue you’re currently facing? This will help us track and assist you more effectively.

If this has addressed your original question, I would appreciate it if you could mark this response as the accepted answer.

Thank you for reaching out to The Microsoft Q&A Portal.

Answer 2

Hi, I finally found a solution. I don't know why but the default value for Instance count is 3 but considering that NC4as_T4_v3 requires 4 cores, then for 3 instances you will need 12 (even more considering 20% quota than expected may be redundancy purposes on some SKUs). Due to the default value of 3 my quota of 8 did not cover that many cores, so NC4as_T4_v3 did not even appear in the VM list.

My advice is always set the instance to 1 to see all available VMs. Also consider the quota and those same 20%. Good luck

Share via

Cannot use GPU VMs (NC4as_T4_v3) in Azure ML despite having quota - VMs grayed out

Subscription Details

Quota Status

Problem Description

What I've Verified

My Question

2 answers

Your answer