This document describes Compute Engine instances that use thereservation-bound provisioning model, including their benefits and creationrequirements.

When you create an instance, you must also specify the underlyingprovisioning model, which definesthe availability, pricing, and lifespan for your instances. Thereservation-bound provisioning model lets you create A4X Max, A4X, A4, or A3Ultra instances, both bare metal and virtual machine instances, by usingreserved capacity from afuture reservation in calendar modeorfuture reservation in AI Hypercomputer.

The reservation-bound provisioning model offers the following benefits:

• Cost control: you don't incur additional charges when you createinstances by using your reserved capacity. You only incur charges forresources that aren't part of your reservation, such as disks or IPaddresses.

• Lifecycle management: based on the termination action that you specifywhen you create instances, Compute Engine stops or deletes theinstance at the end of the reservation period.

Understand instances that use the reservation-bound provisioning model

The following sections describe the requirements that you apply when you createcompute instances by using the reservation-bound provisioning model.

Instance creation prerequisites

To use the reservation-bound provisioning model to create compute instances, youmust first reserve resources. You can reserve resources as follows:

• To reserve up to 80 A4 or A3 Ultra instances, and for up to 90 days, usefuture reservation requests in calendar mode.

• To reserve more than 80 A4X Max, A4X, A4, or A3 Ultra instances, and forlonger than 90 days, usefuture reservation requests in AI Hypercomputer.

If Google Cloud approves your future reservation request, thenCompute Engine automatically creates (auto-creates) a reservation atthe start of your reservation period. You can then use the reservation to createinstances.

Instance creation requirements

To create a compute instance by using the reservation-bound provisioning model,you must specify the following configurations:

• The instance and the reservation must have matching properties. You canonly use your reserved capacity to create instances if the instance andauto-created reservation propertiesexactly match. For more information,see therequirements for consuming reservations.

• The instance must specifically target the reservation for consumption.When you create an instance, you must specify the name of the auto-createdreservation to target for consumption, as well as set thereservationAffinity field toSPECIFIC_RESERVATION. For more information,seeConsume a specifically targeted reservation.

• The instance must use the reservation-bound provisioning model. When youcreate an instance, you must specify the reservation-bound provisioningmodel as follows:

  • In the Google Cloud console, in theProvisioning model list, selectReservation-bound.

  • In the Google Cloud CLI, include the--provisioning-model=RESERVATION_BOUND flag in the command.

  • In the Compute Engine API, include the"provisioningModel": "RESERVATION_BOUND" field in the request body.

• The instance must be stopped or deleted at the reservation end time.When you create an instance, you must specify whether to stop or delete theinstance at the reservation's end time by using theinstanceTerminationAction field. For more information, see how tolimit the run time of an instance.

After you create the instance, the instance starts running and keeps runninguntil you stop or delete it, or until the Compute Engine stops ordeletes the instance at the reservation's end time.

Quota

When you create a compute instance by using the reservation-bound provisioningmodel, you don't need quota for the reserved resources that you use to createthe instance. You only need quota for the resources that aren't part of yourreservation, such as disks and IP addresses. For more information about thedifferent types of quota, seeAllocation quotas.

Pricing

When you create a compute instance by using the reservation-bound provisioningmodel, you incur charges as follows:

• Charges start when you create the instance. However, you don't incuradditional charges for the reserved resources that you use to create yourinstance. You only incur charges for the resources that aren't part of thereservation, such as disks or IP addresses. For more information, see thebilling for reservations.

• Charges stop when the reservation that you used to create the instancereaches its end time. At that time, Compute Engine deletes thereservation, and stops or deletes your instance based on the terminationaction that is specified in the instance.

  Important: After Compute Engine stops an instance and changes itsstate toSTOPPING, you continue to incur charges for any resources thatare attached to the instance, such as disks or GPUs. To avoid unnecessarycosts, detach and delete any resources that you no longer need, or deletethe instance altogether. For more information, see thepricing for an instance's uptime.

Limitations

Compute instances that you create by using the reservation-bound provisioningmodel can only use the following machine series:

• A4X Max

• A4X

• A4

• A3 Ultra

To inquire about using other accelerator-optimized machine series with thereservation-bound provisioning model, contact youraccount team or thesales team.

What's next

• Create a compute instance that uses the reservation-bound model