An Azure machine learning service for building and deploying models.
No Azure region can be guaranteed to have less GPU traffic; shortlist regions where your required GPU SKU is available to your subscription, check regional quota, and then request quota and create a new Azure Machine Learning compute cluster in the best candidate region.
1. Check the exact GPU SKU across candidate regions
Quota is regional and VM-family-specific, and approval is not a guarantee of real-time physical capacity.^1^
Use Azure CLI to check whether your required SKU is offered or restricted for your subscription. Replace the example SKU and add regions that satisfy your latency and data-residency requirements:
sku="Standard_NC4as_T4_v3"
for region in eastus2 centralus southcentralus westus3; do
echo "=== $region ==="
az vm list-skus \
--location "$region" \
--size "$sku" \
--resource-type virtualMachines \
--all \
--output table
done
The --all option includes restricted SKUs. A SKU reported as NotAvailableForSubscription is not currently usable by that subscription in that region.^2^
This check establishes SKU support and subscription restrictions—not current allocatable GPU inventory.
2. Compare quota in the remaining regions
In the Azure portal:
- Open Subscriptions and select your subscription.
- Select Usage + quotas.
- Filter by each candidate Region and the required GPU VM family, such as
NCasT4_v3. - Check both:
- The GPU VM-family vCPU quota.
- The total regional vCPU quota.
- Submit a quota request for a candidate region that exposes the required family.
Azure Machine Learning has per-region limits for VM-family cores, total cores, and managed compute resources. GPU families commonly begin with zero cores, so moving regions does not automatically provide usable quota.^1^
The requester needs suitable permissions. The built-in Quota Request Operator role can read and create quota requests, retrieve their status, and create related support tickets; otherwise, ask the subscription administrator to submit the request.^3^
3. Check subscription eligibility
If this is a Free, Trial, benefit, or sponsorship subscription, quota increases may be unavailable for the offer. In the matching Azure Machine Learning case, free subscriptions were not eligible for quota increases and the supported path was upgrading to Pay-As-You-Go.^4^
Check Subscriptions → your subscription → Overview for the offer type before spending time testing more regions. Quota defaults also vary by subscription offer category.^1^
4. Create cross-region Azure ML compute
For an Azure Machine Learning compute cluster:
- Open Azure Machine Learning studio.
- Go to Manage → Compute → + New → Compute cluster.
- In Location, select the region where the quota was approved.
- Select GPU, then choose the approved VM size.
- Configure the node limits and select Create.
Azure ML explicitly supports choosing a compute-cluster region different from the workspace region when the default region lacks quota.^5^
This documented cross-region procedure applies to compute clusters. If the requirement is specifically a compute instance, do not assume the same location workflow; verify whether Studio offers the required location during creation. A compute instance’s size cannot be changed after creation, and its ability to start remains dependent on regional capacity.^6^
Cross-region compute can add network latency and outbound data-transfer cost, particularly when the workspace, datastore, or virtual network remains in East US.^5^
If requests for multiple eligible regions are rejected, open an Azure quota support request and ask for a review of the subscription’s GPU quota eligibility and regional options. Support cannot guarantee that a particular region or GPU family will be approved or have immediate capacity.
References
- Manage and increase quotas and limits for resources with Azure Machine Learning
- Resolve errors for SKU not available
- Azure Quota API
- can't create GPU instance & can't request quota in Machine learning - Microsoft Q&A
- Create an Azure Machine Learning compute cluster
- Create an Azure Machine Learning compute instance