Bring Your Own Reservations
Learn how to use Bring Your Own Reservations (BYOR) with Data Science workloads.
Overview
Learn which Data Science workloads support Bring Your Own Reservations (BYOR) and which aren't supported.
BYOR lets you use your own Oracle Cloud Infrastructure (OCI) Compute Capacity Reservations when running Data Science workloads.
BYOR supports the following Data Science resources:
- Single-node Jobs and Job Runs
- Notebook Sessions
- Model Deployments
- AI Quick Actions model deployments
BYOR supports Notebook Sessions, single-node Jobs and Job Runs, Model Deployments, and AI Quick Actions model deployments. BYOR doesn't support Pipelines or multi-node Job Runs.
BYOR helps you use reserved Compute capacity, including GPU shapes, without relying on service-managed reservations.
Limits
Capacity reservations don't replace Data Science service limits.
Data Science limits continue to control the maximum amount of reservation-backed capacity that workloads can consume within a compartment. Administrators can use these limits to control reservation usage across teams and projects.
Billing and Security
You're charged according to the pricing associated with your capacity reservation.
All workloads remain subject to the IAM and compartment controls configured in your tenancy.
For more information, see the Data Science documentation.
Using Bring Your Own Reservations
Use a capacity reservation when you create or update a supported Data Science workload in the Oracle Cloud Console.
Before you use Bring Your Own Reservations (BYOR), ensure that the following requirements are met:
- You have a Compute Capacity Reservation for the required shape in a valid OCI compartment.
- You have the required reservable capacity limits for the shapes that you want to reserve.
- You have the required IAM policies.
- You have the OCID of the capacity reservation that you want to use.
- You can create and manage Data Science resources in the target compartment.
- Raise a request with Oracle to enable your tenancy and the target region for the BYOR feature.
For more information about creating and managing capacity reservations, see Compute Capacity Reservations.
The following table maps supported GPU shapes to the required reservable limits:
| Shape Family | GPU Shape | Reservable Limit |
|---|---|---|
| A10 | VM.GPU.A10.X | gpu-a10-core-count-reservable |
| A10 | BM.GPU.A10.4 | gpu-a10-core-count-reservable |
| A100 (E3) | BM.GPU4.8 | gpu4-core-count-reservable |
| A100 (B4) | BM.GPU.B4.8 | gpu-b4-core-count-reservable |
| A100 v2 | BM.GPU.A100-v2.8 | gpu-a100-v2-core-count-reservable |
| L40S | BM.GPU.L40S-NC.4 | gpu-l40s-nc-core-count-reservable |
| H100 | BM.GPU.H100.8 | gpu-h100-core-count-reservable |
| H200 | BM.GPU.H200.8 | gpu-h200-core-count-reservable |
| GPU2 | VM.GPU2.1 | gpu2-core-count-reservable |
| GPU3 | VM.GPU3.X | gpu3-core-count-reservable |
For information about service limits, see Compute service limits.
To use the API, specify the capacity reservation OCID in the appropriate request field for the workload. See the following API documentation:
For more information, see Data Science REST API Reference.
To use the CLI or an SDK, use a version that supports the BYOR request fields. For more information, see CLI Reference for Data Science.
To use BYOR from the Console, do the following:
- Create or update a supported Data Science workload.
- Select the target compartment.
- Select the capacity reservation that you want to use.
- Complete the remaining configuration.
- Create or update the workload.
Supported workloads include:
- Single-node Jobs and Job Runs
- Notebook Sessions
- Model Deployments
- AI Quick Actions model deployments
For AI Quick Actions, BYOR applies only when the workflow creates a model deployment. During the model deployment step, select the capacity reservation that you want to use. For more information, see AI Quick Actions.
See the following documentation:
Migrating from Service Managed Reservations
Migrate supported workloads from service-managed reservations to Bring Your Own Reservations (BYOR).
If you have available reservation capacity, you can begin using customer-managed reservations immediately.
Migrating workloads
Complete all BYOR prerequisites before beginning migration.
If reservation capacity is limited, migrate workloads in phases.
- Stop existing service-managed workloads.
- Migrate workloads in phases.
- Validate reservation availability after each migration.
- Coordinate migration planning with the service team as needed.
Frequently Asked Questions
Find answers to common questions about Bring Your Own Reservations (BYOR) support and behavior.
Can I select my reservation in the Console?
Yes. Select the compartment and then select the reservation within that compartment.
Does BYOR work for all workloads?
The following workloads are supported:
- Notebook Sessions
- Single-node Jobs and Job Runs
- Model Deployments
- AI Quick Actions model deployments
What happens if I don't specify a capacity reservation?
The workload uses on-demand pool capacity.
What happens if my reservation doesn't have enough capacity?
Resource creation fails.
Verify available reservation capacity before creating the workload.
Does BYOR support CPU shapes?
Yes.
Are default reservations supported?
No. Default reservations aren't supported currently.
Do I still need to specify a shape?
The selected shape must match a shape that's supported by the specified reservation.