Overview¶

Cloud-Native Support¶

The RBLN SDK is designed so that the RBLN NPU family can be used consistently in container-based and Kubernetes-based cloud-native AI environments. From a container on a single server to a large Kubernetes cluster or an enterprise OpenShift platform, the same software configuration makes NPUs available and lets you deploy and operate AI workloads.

RBLN NPU Operator¶

Operating special hardware such as NPUs in a Kubernetes cluster requires more than exposing the devices to workloads: software such as drivers, container runtimes, and libraries must be present across all nodes in mutually compatible versions. This configuration must be redone whenever node configuration or versions change, so operation involves repetitive manual work. The RBLN NPU Operator automates the management of this software through the Kubernetes operator framework, so that an administrator can operate the cluster by declaring the desired NPU configuration.

For details, see the RBLN NPU Operator documentation.

RBLN Container Toolkit¶

The RBLN Container Toolkit simplifies building and deploying NPU-based container environments. It follows the Container Device Interface (CDI) specification for portability and injects runtime libraries and NPU devices into containers, which integrates it with container orchestration environments including Kubernetes.

For details, see the RBLN Container Toolkit documentation.

RBLN Cloud Native Stack¶

The RBLN Cloud Native Stack is a reference architecture for configuring the RBLN NPU Operator on upstream Kubernetes. It bundles the components an NPU cluster needs as a validated combination: Kubernetes cluster setup, container runtime configuration, RBLN NPU Operator deployment, and container support for the RBLN SDK. It includes examples for several infrastructure environments, so you can build a validated RBLN NPU-based Kubernetes environment yourself.

The RBLN Cloud Native Stack can be used to confirm functionality and performance with a validated configuration before adopting a new NPU server or SDK version, or as a standard environment for building and testing NPU-accelerated container applications. A configuration validated here can be applied in the same way to enterprise Kubernetes platforms that use the RBLN NPU Operator.

For the configuration and installation procedure, see the RBLN Cloud Native Stack repository. For the supported version of each component, see the Support Matrix.