June
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-xx¶
Mid June, 2025
This are draft release notes for the upcoming release of Rafay GPU PaaS on the Air Gapped Controller.
White Labeling - Branding and Themes¶
Cloud providers can now customize the look and feel of their platform by configuring CSS-based theming. Primary and secondary colors can be set directly via the Ops Console, and these updates are automatically applied across both the PaaS Studio and the end customer-facing Developer Hub (self-service portal), ensuring a consistent branded experience.
vCluster Plugin for KubeVirt based VMs¶
Admins can now avoid installing KubeVirt inside each vCluster. They can leverage the existing KubeVirt installation on the host Kubernetes cluster avoiding duplication of resources. This allows cloud providers to operate both vClusters and VMs on top of a shared host cluster, dramatically simplifying operational burden.
Custom Compute/Service Types¶
Cloud providers can now define and assign custom compute and service classifications (e.g. baremetal, vm for "compute type" AND notebook, inference etc for "service type"). To enable this, we have introduced a new field "computeType" and "serviceType" in the declarative specs for the compute and service profile resources.
New View in End User Self Service Portal¶
End users are now presented with a very intuitive view by type in their Self Service Portal (i.e. Developer Hub). In the example below, you can see the default view for end users before and after this update.
Default Types¶
The following types are supported out of box and are available for use by PaaS Studio administrators that create and curate SKUs.
Compute
- Bare Metal
- VM
- vCluster
- Kubernetes
Service
- Notebook
- Inference
- AI/ML Job
- Custom Services
Select Type during SKU Creation¶
Users in the PaaS Studio can select the type when they create a SKU. In the example below, the admin is presented with a dropdown selector for type when they create a profile.
Note
The compute/service type cannot be updated after the profile has been created and saved.
View by Type and Sort¶
PaaS Studio administrators using the PaaS Studio and end users can view the type for every compute and service profile. They can also sort and view the profiles by clicking on the column header. In the example below, there are a number of compute profiles of various types (Bare Metal, VM, vCluster etc).
Overrides¶
The defaults provided by Rafay may not work for all cloud providers. They may wish to create new types/categories for their profiles. Administrators can perform this using the Operations Console as in the example below.
Profile Category¶
Administrators can now specify annotations to profiles that will make sure that it will be automatically listed under a category in the tenant's profile catalog. For example, if the user adds and updates the annotation value (e.g., sets it to AI/ML), the profile should then appear under the AI/ML category in the tenant organization's Profile Catalog.
Centralized Identity Provider¶
Cloud Providers can deploy and operate a centralized Identity Provide (IdP) to provide a single sign on (SSO) experience for all their users spanning multiple customer orgs (i.e. tenants). This dramatically simplifies user onboarding, reduces IdP sprawl and centralizes user management.
Important
The users will be dynamically routed to the correct org based on metadata from the IdP (e.g., from the SAML response).
Project and User Quotas¶
Resource usage is governed by a hierarchical quota model to ensure fair and controlled allocation across different organizational scopes. The cloud provider will set quota at the "Org" level. This is a global cap for the number of instances of SKUs that can be launched across the Org. The administrator for the customer Org can set quotas for downstream users using the approaches described below.
Project-Level and User Level Quota¶
Customer Org Admins can use this to set quotas at the project level. For example, if they set a quota of 5 for a SKU, no more than 5 instances of the SKU can be launched under that specific project, even if the organization quota has remaining capacity. This ensures that org level capacity can be equitably shared by users spanning multiple projects.
Within each project, administrators can assign quota for users assigned to that project on a per SKU (profile) basis. For example, if a user’s quota is 1, any user in the project can launch only one instance under that project.
