Azure container app - Consuption profile + GPU - COST CALCULATION

Hello @Fatehali Sunasara,

Azure Container Apps with the Azure Container Apps Consumption-GPU workload profile is billed using a pay-as-you-go model, so the cost is calculated based on actual resource usage per second rather than a fixed monthly price. This is why the Azure Pricing Calculator currently does not provide a direct option to estimate costs for GPU-enabled Container Apps.

How the cost is calculated

For GPU workloads, the total cost is the sum of the following components:

• GPU usage (per second)
• vCPU usage (per second)
• Memory usage in GiB (per second)
• Requests processed (after the free tier)

In simplified form:

Total Cost =

With the Consumption-GPU-NC8as-T4 workload profile, a replica can use approximately:

• 1 NVIDIA T4 GPU
• Up to 8 vCPUs
• Up to 56 GiB memory

Billing occurs only while a replica is running. If the container app scales to zero, no compute charges are incurred during that time.

Example estimation

If one replica runs 8 hours per day for 30 days:

Total hours = 8 × 30 = 240 hours

Approximate cost estimation (varies by region):

• GPU ≈ $0.90–$1.20 per hour
• CPU + Memory ≈ $1.20 per hour (approximate combined cost)

Estimated hourly cost ≈ $2.1–$2.4 per hour

Monthly estimate:

240 hours × ~$2.3/hour ≈ $550/month

Key factors affecting cost

• Replica count – multiple replicas multiply the cost.
• Minimum replicas setting – if minReplicas = 1, the GPU runs continuously.
• Scaling behavior – scaling to zero significantly reduces cost.
• Region pricing – GPU pricing varies by region.

Recommendation

Since the pricing calculator does not yet support GPU consumption profiles for Azure Container Apps, the best approach is to:

1. Estimate expected runtime hours per month
2. Multiply by approximate hourly cost of GPU + CPU + memory
3. Adjust for number of replicas

You can also monitor actual costs using Cost Analysis in the Azure portal after deployment

Please accept as answer and do a Thumbs-up to upvote this response if you are satisfied with the community help. Your upvote will be beneficial for the community users facing similar issues.

Hi @Fatehali Sunasara

Thank you for reaching out to Microsoft Q&A.

The issue you’re encountering is a current limitation of the Azure Pricing Calculator for Azure Container Apps. When using Azure Container Apps with a Consumption workload profile and GPU (for example, Consumption‑GPU‑NC8as‑T4), the service is billed using serverless, per‑second meters (vCPU‑seconds, GiB‑seconds, and GPU‑seconds). However, the Azure Pricing Calculator does not yet expose a GPU option for Azure Container Apps, so there is no way to directly select a GPU profile or see GPU pricing in the calculator UI. Because of this, a fixed “monthly cost” cannot be pre‑calculated from the portal, as the actual charge depends entirely on how long your replicas run and how much GPU time they consume.

Refer below points to resolve this issue or as a workaround

Understand how GPU billing works in Consumption profile Azure Container Apps with serverless GPUs are billed per second while replicas are running. Charges include vCPU‑seconds, memory (GiB‑seconds), and GPU‑seconds. When the app scales to zero, GPU billing stops. There is no flat monthly price.

Estimate cost using expected GPU usage (manual calculation) As a workaround, estimate the number of hours your app will actively use the GPU per month (for example, replicas × hours per day × days per month). Use the hourly price of an equivalent NVIDIA T4 GPU (such as NCas T4 VM pricing in your region) as a reference and multiply it by the estimated GPU hours. This gives an approximate monthly GPU cost, to which CPU and memory consumption charges are added.

Use Azure Cost Management for accurate values For the most accurate numbers, deploy the workload and review Azure Cost Management → Cost by meter. Look specifically for GPU‑seconds, vCPU‑seconds, and GiB‑seconds. This is currently the only authoritative way to see the exact cost for Azure Container Apps with GPUs.

Plan scaling carefully to control cost Configure minimum and maximum replicas appropriately and allow scale‑to‑zero where possible. Since GPU costs are incurred only while replicas are running, optimizing scaling behavior is the most effective way to control monthly spend.

Azure Container Apps GPU on the Consumption-GPU-NC8as-T4 profile is billed as:

1. A Consumption plan charge for vCPU, memory, and GPU usage per replica while the app is running.
2. Any applicable Dedicated Plan Management fee if using features such as private endpoints or planned maintenance.

However, the exact per‑vCPU, per‑GiB, and per‑GPU‑hour prices for Consumption-GPU-NC8as-T4 are not listed in the provided documentation and the Azure Pricing Calculator currently doesn’t expose a dedicated GPU toggle for Container Apps.

To get the monthly cost for this profile:

1. Go to the Azure pricing calculator for Azure Container Apps.
2. Sign in so negotiated prices (if any) are applied.
3. Configure:
   • Plan type: Consumption.
   • vCPUs / memory: Match the expected average usage of the app (up to 8 vCPUs, 56 GiB for this GPU profile, but enter what is actually used and for how many hours per month).
   • Replica count and hours per month: Estimate how many replicas run concurrently and for how long.
4. For the GPU component, use the same calculator page once GPU pricing for Container Apps is exposed in your region, or cross‑check the Azure Container Apps Pricing page from the billing article to see the per‑hour rate for serverless GPUs, then multiply: Monthly GPU cost ≈ GPU price per hour × GPU hours used per month
5. Add any Dedicated Plan Management fee if you use features like private endpoints or planned maintenance, as noted in the billing article.

Because GPU serverless pricing is region‑specific and not detailed in the context, the authoritative way to get the exact monthly estimate is to:

• Use the Azure pricing calculator for Container Apps, and
• Check the Azure Container Apps Pricing page linked from the billing documentation for the GPU per‑hour rate in the target region, then apply the formula above based on expected usage (replicas × hours).

References:

• Workload profiles in Azure Container Apps
• Provision Azure Container Apps
• Compute and billing structures in Azure Container Apps
• Tutorial: Generate images using serverless GPUs in Azure Container Apps (azure-portal)
• Billing in Azure Container Apps
• Plan to manage Azure costs