Plan and manage an Azure AI solution 1.3 Manage, monitor, and secure AI systems

A team plans a steady, long-running production workload on a provisioned (PTU) deployment…

Study guide skill: Manage quotas, scaling, rate limits, and cost footprints for model and agent workloads

Question B8-006

A team plans a steady, long-running production workload on a provisioned (PTU) deployment and wants the lowest effective cost. Which two statements are correct? (Choose two.)

Select 2 answers.

  • A Stay on hourly billing and scale the PTU count up and down with traffic to minimize cost
  • B Buying a reservation guarantees that capacity will be available in the region
  • C Provisioned deployments are billed per token consumed, so idle PTUs cost nothing
  • D Create the deployment first to confirm capacity, and then purchase the reservation, because reservations don't guarantee capacity ✓ Correct answer
  • E A reservation is a discount on the PTU billing meter, bought per deployment type, rather than being attached to one specific deployment ✓ Correct answer

Answer: D, E

Reservations are a financial discount on the PTU billing meter, purchased per deployment type and loosely coupled to deployments, and they don't guarantee capacity, so you create deployments first and then buy the reservation. Provisioned deployments are billed hourly per deployed PTU regardless of tokens consumed. The docs advise against scaling up and down to stay on hourly billing, because continuous hourly billing at high utilization typically exceeds reservation pricing.

Deep dive

How it works

Provisioned (PTU) deployments support two billing modes: hourly billing for flexible, short-term usage, and Azure Reservations for sustained production workloads at a discounted rate. A steady, long-running workload is the reservation case.

Hourly billing. All provisioned deployment types are billed hourly per deployed PTU, regardless of the number of tokens consumed. Idle PTUs still cost money.

Reservations. They're a financial discount applied to the PTU billing meter, not to individual deployments. Key points from the page:

  • You commit for 1 month or 1 year in exchange for a discounted effective $/PTU/hr rate.
  • Reservations are purchased per deployment type (Global, Data Zone or Regional) and can cover one or more subscriptions or resource groups.
  • Reservations and deployments are loosely coupled: you create them independently.
  • Reservations don't guarantee capacity. First create the deployments to confirm capacity is available, then buy the reservation to lock in the discount.

The other options, and when you'd use them

  • Stay hourly and scale with traffic: Microsoft says not to plan on this. Capacity might not be available when you need to scale back up, and continuous hourly billing at high utilization typically costs more than reservation pricing.
  • A reservation guarantees capacity: It doesn't. Having PTU quota doesn't guarantee capacity either.
  • Billed per token, idle is free: Provisioned deployments are billed per deployed PTU per hour, regardless of tokens consumed.

Exam tip

PTU: deploy first, then reserve. Reservation = discount on the meter, per deployment type, no capacity guarantee. Billing is per PTU-hour, not per token.

References

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Plan and manage questions · AI-103 study guide

Thanks for your feedback!