Dec
Info
GPU PaaS releases are initially rolled out via Rafay's Air Gapped Controller form factor. These will be periodically bundled and rolled out into Rafay's Production SaaS.
v3.1-38¶
19 Dec, 2025 (est.)
Inventory: Bulk Node Onboarding¶
For VM as a Service SKUs, admins now have access to a "bulk node onboard" workflow to manage their server inventory. This allows admins to onboard 10s or 100s of servers at the same time enabling providers to add capacity quickly and efficiently.
Inventory: MIG Support¶
Fractional GPUs based on Nvidia MIG are now supported natively in the inventory.
Inventory: Kubernetes Cluster¶
Inventory based on Kubernetes clusters will be automatically reconciled and kept updated based on current usage on clusters.
SKU Mgmt: Configurable Controls for SKUs¶
We have introduced configurable automation and policy controls for how SKUs are shared with tenant orgs and projects within these orgs.
Controls¶
| Setting | Description |
|---|---|
| Share With Specific Org | SKU is available to specified tenant org only |
| Share With All Projects | SKU is made available to all projects within each org |
Behavior Matrix¶
| Org Scope | Auto Share with All Projects | Result / Behavior |
|---|---|---|
| All | True | SKU is shared with all tenant orgs and all projects within each org |
| All | False | SKU is shared with all orgs, but not automatically shared with their projects. The tenant admin controls project-level sharing. |
| Tenant-X | True | SKU is shared with tenant-x and all projects within that tenant. |
| Tenant-Y | False | SKU is shared with tenant-y, but not automatically shared with its projects. |
SKU Mgmt: Per SKU, Per Tenant Configuration¶
When a SKU is shared with all tenant orgs, it is currently not possible to hide it from a specific org.
In this release, we have introduced a "per-SKU" & "per-Tenant" configuration to disable a SKU in an Org. When enabled, the tenant org will no longer see the SKU/profile
Info
Existing instances based on the SKU will not be impacted and can still be viewed or used.
Customization of Welcome Email for New Users¶
Service Providers can now customize the welcome email content that is sent to new users of the platform. For example, the provider may wish to provide links to a user guide in the welcome email.
Usage Reports¶
Cloud Providers can now generate and download usage reports for their end customers (tenant organizations). This capability is available in the default Org and enables Cloud Providers to:
- Select a target Org
- Filter usage by a specified date range
- Include additional billable columns (e.g., storage or other parameter values)
- Generate reports in CSV format
Two types of usage reports are supported:
- Detailed Report – Hourly usage granularity
- Summary Report – Aggregated usage overview
Tenant Dashboard Improvements¶
A number of enhancements have been introduced to improve clarity and usability across the Tenant Dashboard and Developer Hub:
- Display Name Support for SKUs: Added support for display names for compute and service SKUs (previously referred to as profiles).
- SKU Label Standardization: All references to “Profile” have been updated to “SKU” across the Tenant Dashboard and tDev Hub. Compute and service profiles are now consistently represented as Compute SKUs and Service SKUs.
- Current Utilization Fixes: Resolved issues in the Current Utilization section. The type filter has been moved into the GPU card to reflect that it applies only to GPU metrics.
- Project Visibility Enhancements: SKU drawers now show the parent project where each SKU was originally created, along with all projects where the SKU is shared—across both Tenant and Partner Orgs via the “All” tab.
- Improved SKU Ordering: SKUs are now sorted in ascending order during instance creation flows for compute and service workloads in both PaaS Studio and Dev Hub.
SKU Studio¶
The PaaS Studio portal used by admins to design and manage SKUs has been renamed as "SKU Studio".
- Only GPU Cloud Providers (not their end customers) will have access to this portal.
- PaaS Admin & Org Admin in the provider's "Default Org" will have access to SKU Studio.
New Org Type¶
Service Providers can now create Orgs with the type "CSP". For users in these orgs, the Infra and SKU Studio portals are not available/visible.
New Roles for CSP Customer Org Type¶
In this release, we are introducing new roles for users (admins and users) in Tenant Orgs to streamline their day-to-day operations.
Read Only Collaborator¶
End users can now invite collaborators to their workspace with a "read only" role. These users can view/use the compute instances and services, but cannot modify (i.e. update, delete) them.
Tenant Admin¶
A new role for customer admins instead of the Org Admin role. Tenant Admins can perform the following tasks in their Orgs.
- Create projects for teams
- Share CSP's profiles with projects
- Customize profiles with overrides (if allowed by the GPU Provider)
- Set/Manage Quota for Projects
- Configure Policies (e.g. Schedule)
- Manage Users
- Configure Identity Provider for Single Sign On (SSO)
- View Audit Logs
- Manage Other System Settings
Tenant Project Admins¶
A new administrator role scoped to a specific project (e.g. manager of a AI/ML team) enabling delegated administration.
- Add/remove existing IDP users or local users to their project
- View Number of running instances
- View SKUs shared with the project
- Apply quotas and policies
- List of users with access and the instances they are running
- Visualize Cost of running instances
- View instance utilization trends
Managed Kubernetes¶
Control plane and worker nodes can now be provisioned on RHEL 10 and Flatcar Linux.
Managed Kubernetes Versions¶
Support for new cluster provisioning of Kubernetes v1.33 and v1.34. Existing k8s clusters can be upgraded in-place to these versions. The default k8s version with this release is v1.33
Info
k8s v1.34 brings support for DRA. Read our blogs about this new feature for GPUs
External Storage for Controller¶
During installation of the self hosted/air gapped controller, admins can now specify a CSI provider for external storage managed by the organization. For example, the organization may have standardized on a storage platform such as Vast or DDN or Weka or Ceph. This is the recommended and required option for production deployments.
Info
The default "openebs" based CSI option is still available, but limited to only POCs and Demos.
Internationalization & Localization¶
The end user facing self service portal can now be localized to support different languages via language resource bundle files provided via the Ops Console. Shown below is an example for the user would see when they select Japanese as their preferred language.
Shown below is an example in Turkish for what the user would see when they try to launch a SKU.
Info
As of this release, we support only left-to-right languages, with additional language support planned for the future.