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 describes the supported node sizes and scale requirements for AKS on Azure Local for multi-rack deployments.
Support count for AKS on Azure Local
The table below shares minimum count supported. While there are no enforced maximums for clusters per deployment, the effective maximum number of nodes and clusters are constrained by the available physical compute, memory, and logical network IP capacity in your deployment. Plan your logical network IP pools to have enough addresses for all cluster nodes, control plane nodes, plus two more IPs per cluster for internal services.
| Scale item | Minimum |
|---|---|
| Count of control plane nodes (odd numbers) | 1 |
| Number of node pools in an AKS cluster | 1 |
| Number of nodes in a node pool (empty node pools not supported) | 1 |
Note
The control plane node count must be an odd number (>=1) to maintain etcd quorum.
Autoscaler
Cluster autoscaler isn't currently supported on Azure Local for multi-rack deployments. You can manually scale node pools by updating the node count.
Default values for virtual machine sizes
| System Role | VM Size | Memory, CPU |
|---|---|---|
| AKS Arc control plane nodes | Standard_D4s_v3 | 16-GiB memory, 4 vCPU |
| AKS Arc Linux worker node | Standard_A4_v2 | 8-GiB memory, 4 vCPU |
Supported values for control plane node sizes
| VM Size | CPU | Memory (GiB) |
|---|---|---|
| Standard_D4s_v3 | 4 | 16 |
| Standard_D8s_v3 | 8 | 32 |
Supported values for worker node sizes
| VM Size | CPU | Memory (GiB) |
|---|---|---|
| Standard_A2_v2 | 2 | 4 |
| Standard_K8S3_v1 | 4 | 6 |
| Standard_A4_v2 | 4 | 8 |
| Standard_D4s_v3 | 4 | 16 |
| Standard_D8s_v3 | 8 | 32 |
| Standard_D16s_v3 | 16 | 64 |
| Standard_D32s_v3 | 32 | 128 |
GPU requirements
Currently, the only supported GPU model is the NVIDIA RTX Pro 6000 Blackwell (96 GiB). For supported GPU-enabled VM sizes and deployment steps, see Use GPUs for compute-intensive workloads.