How is Managed Compute on Azure AI Studio actually billed?

Question

Can anyone clarify how Managed Compute in Azure AI Studio is billed?

Is managed compute active 24/7 or does it go into a suspended state automatically?

If managed compute is active constantly, this would mean being billed even when there is no usage, and would be very expensive to run for smaller applications.

Answer 1

Managed Compute in Azure AI Studio is billed based on the resources provisioned and the time they are in use. The billing encompasses several components, including compute resources, storage, and networking. Compute resources such as virtual machines and GPUs are billed according to their type, size, and the duration they remain active. Additionally, storage for datasets, models, and other artifacts incurs charges, covering both the storage space and associated read/write transactions. Data transfer costs may also apply, especially if there is significant ingress or egress traffic related to your AI workloads.

When it comes to the compute state, managed compute resources are billed while they are active, which includes both task performance time and idle periods if they are not explicitly stopped. However, the service configuration may include options for auto-scaling and auto-shutdown to manage costs. These features help to suspend or deallocate resources during periods of inactivity, potentially reducing the overall cost.

To manage costs effectively, you can utilize auto-scaling to dynamically match compute resources to workload demands, avoiding over-provisioning and reducing expenses during low activity periods. Setting up auto-shutdown policies can also help by automatically deallocating resources when they are not in use, which is particularly beneficial for development and testing environments that do not require 24/7 operation.