What is provisioned throughput?

The Azure AI Foundry provisioned throughput offering is a model deployment type that allows you to specify the amount of throughput you require in a model deployment. Azure AI Foundry then allocates the necessary model processing capacity and ensures it's ready for you. You can use the provisioned throughput you requested across a diverse portfolio of models that are sold directly by Azure. These models include Azure OpenAI models and newly introduced flagship model families like Azure DeepSeek, Azure Grok, Azure Llama, and more within Azure AI Foundry Models.

Provisioned throughput provides:

• A boarder model choice on the latest flagship models
• Flexibility to switch models and deployments with given provisioned throughput quota
• Significant discounts and the ability to boost your reservation utilization with a more flexible reservation choice
• Predictable performance, by providing stable max latency and throughput for uniform workloads.
• Allocated processing capacity: A deployment configures the amount of throughput. Once deployed, the throughput is available whether used or not.
• Cost savings: High throughput workloads might provide cost savings vs token-based consumption.

When to use provisioned throughput

You should consider switching from standard deployments to provisioned throughput deployments when you have well-defined, predictable throughput and latency requirements. Typically, this occurs when the application is ready for production or is already deployed in production and there's an understanding of the expected traffic. This allows users to accurately forecast the required capacity and avoid unexpected billing. Provisioned Throughput deployments are also useful for applications that have real-time/latency sensitive requirements.

Key concepts

The sections that follow describe key concepts that you should be aware of when using the provisioned throughput offering.

Provisioned Throughput Units (PTU)

Provisioned throughput units (PTU) are generic units of model processing capacity that you can use to size provisioned deployments to achieve the required throughput for processing prompts and generating completions. Provisioned throughput units are granted to a subscription as quota, and used to define costs. Each quota is specific to a region and defines the maximum number of PTU that can be assigned to deployments in that subscription and region.

Cost management under shared PTU reservation

You can use the PTU capability to seamlessly manage costs for Foundry Models under a shared PTU reservation. However, the required PTU units for deployment and throughput performance are dynamically tailored to the chosen models. To learn more about PTU costs and model latency points, see Understanding costs associated with PTU.

Existing PTU reservations are automatically upgraded to empower customers with enhanced efficiency and cost savings as they deploy Foundry Models. For example, suppose you have an existing PTU reservation with 500 PTU purchased. You use 300 units for Azure OpenAI models, and you choose to also use PTU to deploy Azure DeepSeek, Azure Llama, or other models with PTU capability on Foundry Models.

• If you use the remaining 200 PTU for DeepSeek-R1, the 200 PTU share the reservation discount automatically, and your total usage for the reservation is 500 PTU.

• If you use 300 PTU for DeepSeek-R1, then 200 PTU share the reservation discount automatically while 100 PTU exceed the reservation and are charged with DeepSeek-R1's hourly rate.

To learn about saving costs with PTU reservations, see Save costs with Microsoft Azure AI Foundry Provisioned Throughput Reservations.

Deployment types

When you're creating a provisioned deployment in Azure AI Foundry, the deployment type on the "Create Deployment" dialog can be set to the Global Provisioned Throughput, Data Zone Provisioned Throughput, or Regional Provisioned Throughput deployment type depending on the data processing needs for the given workload.

When you're creating a provisioned deployment in Azure AI Foundry via CLI or API, the sku-name can be set to GlobalProvisionedManaged, DataZoneProvisionedManaged, or ProvisionedManaged depending on the data processing need for the given workload.

To adapt the following Azure CLI example command to a different deployment type, update the sku-name parameter to match the deployment type you wish to deploy.

az cognitiveservices account deployment create \
--name <myResourceName> \
--resource-group  <myResourceGroupName> \
--deployment-name MyDeployment \
--model-name gpt-4o \
--model-version 2024-08-06  \
--model-format OpenAI \
--sku-capacity 15 \
--sku-name GlobalProvisionedManaged

Capacity transparency

The models sold directly by Azure are highly sought-after services where customer demand might exceed service GPU capacity. Microsoft strives to provide capacity for all in-demand regions and models, but selling out a region is always a possibility. This constraint can limit some customers' ability to create a deployment of their desired model, version, or number of PTU in a desired region - even if they have quota available in that region. Generally speaking:

• Quota places a limit on the maximum number of PTU that can be deployed in a subscription and region, and doesn't guarantee capacity availability.
• Capacity is allocated at deployment time and is held for as long as the deployment exists. If service capacity isn't available, the deployment fails.
• Customers use real-time information on quota/capacity availability to choose an appropriate region for their scenario with the necessary model capacity.
• Scaling down or deleting a deployment releases capacity back to the region. There's no guarantee that the capacity will be available should the deployment be scaled up or re-created later.

Regional capacity guidance

To find the capacity needed for their deployments, use the capacity API or the Azure AI Foundry deployment experience to provide real-time information on capacity availability.

In Azure AI Foundry, the deployment experience identifies when a region lacks the capacity needed to deploy the model. This looks at the desired model, version, and number of PTU. If capacity is unavailable, the experience directs users to select an alternative region.

Details on the deployment experience can be found in the Azure AI Foundry Provisioned get started guide.

The model capacities API can be used to programmatically identify the maximum sized deployment of a specified model. The API considers both your quota and service capacity in the region.

If an acceptable region isn't available to support the desired model, version, and/or PTU, customers can also try the following steps:

• Attempt the deployment with a smaller number of PTU.
• Attempt the deployment at a different time. Capacity availability changes dynamically based on customer demand and more capacity might become available later.
• Ensure that quota is available in all acceptable regions. The model capacities API and Azure AI Foundry experience consider quota availability in returning alternative regions for creating a deployment.

How can I monitor capacity?

The Provisioned-Managed Utilization V2 metric in Azure Monitor measures a given deployments utilization on 1-minute increments. All provisioned deployment types are optimized to ensure that accepted calls are processed with a consistent model processing time (actual end-to-end latency is dependent on a call's characteristics).

How utilization performance works

Provisioned deployments provide you with an allocated amount of model processing capacity to run a given model.

In all provisioned deployment types, when capacity is exceeded, the API returns a 429 HTTP Status Error. The fast response enables the user to make decisions on how to manage their traffic. Users can redirect requests to a separate deployment, to a standard deployment instance, or use a retry strategy to manage a given request. The service continues to return the 429 HTTP status code until the utilization drops below 100%.

What should I do when I receive a 429 response?

The 429 response isn't an error, but instead, it's part of the design for telling users that a given deployment is fully utilized at a point in time. By providing a fast-fail response, you have control over how to handle these situations in a way that best fits your application requirements.

The retry-after-ms and retry-after headers in the response tell you the time to wait before the next call will be accepted. How you choose to handle this response depends on your application requirements. Here are some considerations:

• You can consider redirecting the traffic to other models, deployments, or experiences. This option is the lowest-latency solution because the action can be taken as soon as you receive the 429 signal. For ideas on how to effectively implement this pattern see this community post.
• If you're okay with longer per-call latencies, implement client-side retry logic. This option gives you the highest amount of throughput per PTU. The Azure AI Foundry client libraries include built-in capabilities for handling retries.

How does the service decide when to send a 429?

In all provisioned deployment types, each request is evaluated individually according to its prompt size, expected generation size, and model, to determine its expected utilization. This behavior is in contrast to standard deployments, which have a custom rate limiting behavior based on the estimated traffic load. For standard deployments, this custom rate limiting behavior can lead to HTTP 429 errors being generated before defined quota values are exceeded if traffic isn't evenly distributed.

For provisioned deployments, we use a variation of the leaky bucket algorithm to maintain utilization below 100% while allowing some burstiness in the traffic. The high-level logic is as follows:

1. Each customer has a set amount of capacity they can utilize on a deployment

2. When a request is made:

   a. When the current utilization is above 100%, the service returns a 429 code with the retry-after-ms header set to the time until utilization is below 100%

   b. Otherwise, the service estimates the incremental change to utilization required to serve the request by combining the prompt tokens, less any cached tokens, and the specified max_tokens in the call. A customer can receive up to a 100% discount on their prompt tokens depending on the size of their cached tokens. If the max_tokens parameter isn't specified, the service estimates a value. This estimation can lead to lower concurrency than expected when the number of actual generated tokens is small. For highest concurrency, ensure that the max_tokens value is as close as possible to the true generation size.

3. When a request finishes, we now know the actual compute cost for the call. To ensure an accurate accounting, we correct the utilization using the following logic:

   a. If the actual > estimated, then the difference is added to the deployment's utilization.

   b. If the actual < estimated, then the difference is subtracted.

4. The overall utilization is decremented at a continuous rate based on the number of PTU deployed.

How many concurrent calls can I have on my deployment?

The number of concurrent calls you can achieve depends on each call's shape (prompt size, max_tokens parameter, etc.). The service continues to accept calls until the utilization reaches 100%. To determine the approximate number of concurrent calls, you can model out the maximum requests per minute for a particular call shape in the capacity calculator. If the system generates less than the number of output tokens set for the max_tokens parameter, then the provisioned deployment will accept more requests.

Provisioned throughput capability for Models Sold Directly by Azure

This section lists Foundry Models that support the provisioned throughput capability. You can use your PTU quota and PTU reservation across the models shown in the table.

The following points are some important takeaways from the table:

• The model version isn't included in this table. Check the version supported for each model when you choose the deployment option in the Azure AI Foundry portal.

• Regional provisioned throughput deployment option varies by region.

• New models sold directly by Azure are onboarded with Global provisioned throughput deployment option first. The Data zone provisioned option comes later.

• PTU are managed regionally and by offer type. PTU quota and any reservations must be in the region and shape (Global, Data zone, Regional) you wish to use.

• Spillover is an optional capability that manages traffic fluctuations on provisioned deployments. For more information on spillover, see Manage traffic with spillover for provisioned deployments (Preview).

Region availability for provisioned throughput capability

Related content

• Learn about the onboarding steps for provisioned deployments
• Provisioned Throughput Units (PTU) getting started guide