Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
This article discusses Azure, machine and storage, networking, and other requirements for Edge RAG.
Important
Edge RAG Preview, enabled by Azure Arc is currently in PREVIEW. See the Supplemental Terms of Use for Microsoft Azure Previews for legal terms that apply to Azure features that are in beta, preview, or otherwise not yet released into general availability.
Resource requirements
To get started with Edge RAG, you need the following Azure and on-premises resources.
Azure resources
Before deploying Edge RAG, make sure you have the following Azure resources and permissions in place:
| Resource | Description |
|---|---|
| Azure subscription | An Azure subscription. |
| Permissions for Azure Kubernetes Service (AKS) enabled by Azure Arc | Permissions to deploy AKS Arc Kubernetes clusters, create node pools, and install extensions. As part of the prerequisites tasks, see Verify contributor role for Edge RAG Preview enabled by Azure Arc. |
| Microsoft Entra ID permissions | - Permissions to create a Microsoft Enterprise Entra application. - Ability to add new or existing Microsoft Entra users and groups to the application. As part of the prerequisites tasks, you configure authentication for Edge RAG Preview enabled by Azure Arc. |
| Transport Layer Security (TLS) termination certificate | A certificate signed by a company-specific certification authority (CA) or a well-known public CA for secure deployments. If you don't provide one, Edge RAG generates a self-signed certificate. We don't recommend using a self-signed certificate for production environments. |
On-premises resources
The following on-premises resources are required to deploy Edge RAG in your environment:
| Resource | Description |
|---|---|
| Azure Local infrastructure | An instance of Azure Local infrastructure, minimum version 2411. |
| AKS Arc cluster on Azure Local | An AKS Arc cluster running on the Azure Local instance. Use GPUs for better performance; include at least three GPU-enabled VMs in the node pool for image and text scenarios. As part of the prerequisites tasks, you prepare AKS cluster on Azure Local for Edge RAG Preview enabled by Azure Arc. |
| Routable, static IP address | One routable, static IP address for the MetalLB load balancer. If MetalLB is already configured with a routable IP, this requirement can be skipped. The IP must be accessible from client machines. As part of the prerequisites tasks, setting up MetalLB is included in the following articles: - Install networking and observability components for Edge RAG Preview enabled by Azure Arc - Configure DNS for Edge RAG Preview enabled by Azure Arc. |
| Network File System (NFS) | An NFS v3.0 or v4.1 containing your on-premises documents or images. See setup guides for Windows Server and Linux. As part of the prerequisites tasks, see Verify NFS server access for Edge RAG Preview enabled by Azure Arc. |
Minimum hardware requirements
The following table lists the minimum hardware requirements for the virtual machines.
| Mode | VM specs & suggested minimum sizes |
|---|---|
| GPU | 3 x GPU-enabled VMs Recommended sizes (choose one based on GPU): - Standard_NC8_A2 - Standard_NC8_A16 3 x CPU VMs - Minimum spec: 8 vCPUs, 32 GB - Recommended size: Standard_D8s_v3 |
For more information, see Resource limits, VM sizes, and regions for AKS on Windows Server.
The following table lists the hardware recommendations for each language model available with Edge RAG.
| Model Name | GPU Support | Minimum VM SKUs |
|---|---|---|
| Microsoft/Phi-3.5-mini-instruct | Nvidia A2 Nvidia A16 |
Standard_NC8_A2 Standard_NC8_A16 |
| mistralai/Mistral-7B-Instruct-v0.2 | Nvidia A2 Nvidia A16 |
Standard_NC8_A2 Standard_NC8_A16 |
If you plan to use a CPU-only setup, review the files size and chunking limitations. See:
Minimum software requirements
The following table lists the minimum software requirements for Edge RAG.
| Component | Minimum requirements |
|---|---|
| VM Operating System | Linux |
| Azure Local version | Azure Local 2411 release |
| Azure CLI | As shipped with Azure Local. Don't update to the latest version of Azure CLI, and use the one that was originally shipped with Azure Local. |
Network requirements
All current Azure Local and AKS on Azure Local requirements.
Supported document formats and size
Edge RAG supports the following capabilities and related file formats:
| Capability | Supported file format |
|---|---|
| Text extraction | PDF, DOCX, TXT, MHTML, MHT, MD |
| Image ingestion | JPG, JPEG, PNG |
With a GPU setup, each individual file can be up to 30 MB. If you're using a CPU-only setup, each individual file can be up to 5 MB.
Document or image file types not listed, like audio and video files, aren't currently supported.
Supported data sources
Edge RAG supports Network File System (NFS) v3.0 and v4.1 with AUTH_SYS authentication as a data source. Kerberos isn't supported.
Supported regions
Edge RAG is supported in the following regions:
- westeurope
- eastus2euap
- eastus
- westus2
- australiaeast
- eastus2
- japaneast
- canadacentral
- uksouth
- centralindia
- koreacentral