|Index|Introduction to SUSE AI Factory with NVIDIA|What are the building blocks of SUSE AI Factory with NVIDIA?
Applies to SUSE AI Factory with NVIDIA

3 What are the building blocks of SUSE AI Factory with NVIDIA?

SUSE AI Factory with NVIDIA consists of a Kubernetes operator and a Rancher UI extension. It lets you install and manage individual applications from AI Library as well as group them into stacks called blueprints.

An image showing a basic structure of SUSE AI Factory with NVIDIA
Figure 3.1: Basic schema of SUSE AI Factory with NVIDIA

3.1 Foundation

The base layer of SUSE AI Factory with NVIDIA is deployed on bare metal or cloud-hosted nodes, preferably deployed by the SUSE Elemental. It consists of the following components:

  • a prescribed configuration of the SLES 16 operating system

  • RKE2 Kubernetes management cluster

  • SUSE Rancher Prime that supports the UI extension and Kubernetes operator code for SUSE AI Factory with NVIDIA

  • optionally, SUSE Security and SUSE Observability work together to provide high security and advanced monitoring capability across all AI workloads

3.2 AI Factory

SUSE AI Factory with NVIDIA consists of a Kubernetes operator and its Rancher UI extension. It serves the following purposes:

  • Discover, install and manage individual applications from the NVIDIA GPU Cloud (NGC) catalog.

  • Create custom blueprints based on customer’s AI application selection.

Both running applications and blueprints can be shared across Rancher-managed clusters.

3.3 Blueprints

Blueprints are published, immutable and version-controlled stacks of applications required for a specific use-case. Because AI workloads require multiple interconnected applications, a Blueprint tightly packages these necessary components into a single unit that can be deployed with one click.

An example of a Blueprint is a RAG (Retrieval-Augmented Generation) workload. It typically consists of the following components:

  • A vector database such as Qdrant or Milvus, to store high-dimensional vector embeddings.

  • An LLM inference engine such as Ollama or vLLM to process both the user’s prompt and the retrieved context to generate the final textual response.

  • A user interface such as Open WebUI to provide an interface for end-users to interact with the pipeline.

SUSE AI Factory with NVIDIA is delivered with a set of prebuilt NVIDIA blueprints. You can also create custom blueprints and select your own set of AI applications.