When you create a Kubernetes cluster in Google Distributed Cloud (GDC) air-gapped, you
create node pools that are responsible for running your container workloads in
the cluster. You provision nodes based on your container workload requirements,
and can update them as your requirements evolve.

GDC provides predefined machine types for your worker
nodes that are selectable when you
[add a node pool](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform/pa-user/manage-node-pools.md.txt#add-a-node-pool).
There are also multiple ways to partition separate GPU instances using the
Multi-Instance GPU (MIG) feature.

This document is for application developers within the application operator
group and IT administrators within the platform administrator group who are
responsible for managing Kubernetes clusters within their organization. For more
information, see
[Audiences for GDC air-gapped documentation](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/resources/audiences.md.txt).

Reference the following sections for available machine types and GPU support.

## Available machine types

GDC defines machine types with some parameters
for a Kubernetes cluster node, which include CPU, memory, and GPU.
GDC has various machine types for different purposes.
For example, clusters use `n2-standard-4-gdc` for general purpose container
workloads. If you plan to run artificial intelligence (AI) and
machine learning (ML) notebooks, you must provision GPU machines, such as
`a2-highgpu-1g-gdc`.

The following is a list of all GDC predefined machine
types available for Kubernetes cluster worker nodes:

| Name | vCPUs | Memory | GPU |
|---|---|---|---|
| n2-standard-4-gdc | 4 | 16G | N/A |
| n2-standard-8-gdc | 8 | 32G | N/A |
| n2-standard-16-gdc | 16 | 64G | N/A |
| n2-standard-32-gdc | 32 | 128G | N/A |
| n2-highmem-4-gdc | 4 | 32G | N/A |
| n2-highmem-8-gdc | 8 | 64G | N/A |
| n2-highmem-16-gdc | 16 | 128G | N/A |
| n2-highmem-32-gdc | 32 | 256G | N/A |
| n2-highcpu-8-gdc | 8 | 8G | N/A |
| n3-standard-4-gdc | 4 | 16G | N/A |
| n3-standard-8-gdc | 8 | 32G | N/A |
| n3-standard-16-gdc | 16 | 64G | N/A |
| n3-standard-32-gdc | 32 | 128G | N/A |
| n3-standard-64-gdc | 64 | 256G | N/A |
| n3-standard-80-gdc | 80 | 320G | N/A |
| n4-highmem-4-gdc | 4 | 32G | N/A |
| n4-highmem-8-gdc | 8 | 64G | N/A |
| n4-highmem-16-gdc | 16 | 128G | N/A |
| n4-highmem-32-gdc | 32 | 256G | N/A |
| n4-highmem-48-gdc | 48 | 384G | N/A |
| n4-highmem-64-gdc | 64 | 512G | N/A |
| n4-highmem-80-gdc | 80 | 640G | N/A |
| n4-megamem-64-gdc | 64 | 896G | N/A |
| a2-highgpu-1g-gdc | 12 | 85G | 1x A100 40GB |
| a2-ultragpu-1g-gdc | 12 | 170G | 1x A100 80GB |
| a2-ultragpu-2g-gdc | 24 | 340G | 2x A100 80GB |
| a3-highgpu-1g-gdc | 28 | 240G | 1x H100 94GB |
| a3-highgpu-2g-gdc | 56 | 480G | 2x H100 94GB |
| a3-highgpu-4g-gdc | 112 | 960G | 4x H100 94GB |
| a3-ultragpu-8g-gdc | 224 | 1792G | 8x H200 141GB |
| a4-ultragpu-8g-gdc | 224 | 3584G | 8x B300 |

> [!WARNING]
> **Preview:** The following machine types are in Preview:
>
> - a3-highgpu-1g-gdc
> - a3-highgpu-2g-gdc
>
> For more information on Preview features, see [Feature stages](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/resources/feature-stages.md.txt#preview).

## Supported MIG profiles

This section defines the supported partitioning schemes of MIG profiles on
supported GPUs. You can define a partitioning scheme for a node pool in your
`Cluster` custom resource.

> [!IMPORTANT]
> **Important:** A partitioning scheme gets applied to all GPUs in a node. For example, the `a3-highgpu-4g-gdc` machine can support four iterations of the `7x 1g.12gb` GPU slicing because there are four GPUs available to the machine type.

For more information on how to apply a GPU partitioning scheme, see
[Add a node pool](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform/pa-user/manage-node-pools.md.txt#add-a-node-pool).

### A100 40GB GPU

The following table defines the MIG profiles supported on the A100 40GB NVIDIA
GPU:

| Partitioning Scheme | Available Partitions |
|---|---|
| 1g.5gb | 7x 1g.5gb |
| 2g.10gb | 3x 2g.10gb |
| 3g.20gb | 2x 3g.20gb |
| 7g.40gb | 1x 7g.40gb |
| mixed-1 | 1x 4g.20gb 1x 2g.10gb 1x 1g.5gb |
| mixed-2 | 1x 4g.20gb 3x 1g.5gb |
| mixed-3 | 1x 3g.20gb 2x 2g.10gb |
| mixed-4 | 1x 3g.20gb 1x 2g.10gb 2x 1g.5gb |
| mixed-5 | 1x 3g.20gb 4x 1g.5gb |
| mixed-6 | 3x 2g.10gb 1x 1g.5gb |
| mixed-7 | 2x 2g.10gb 3x 1g.5gb |
| mixed-8 | 1x 2g.10gb 5x 1g.5gb |

### A100 80GB GPU

The following table defines the MIG profiles supported on the A100 80GB NVIDIA
GPU:

| Partitioning Scheme | Available Partitions |
|---|---|
| 1g.10gb | 7x 1g.10gb |
| 2g.20gb | 3x 2g.20gb |
| 3g.40gb | 2x 3g.40gb |
| 7g.80gb | 1x 7g.80gb |
| mixed-1 | 1x 4g.40gb 1x 2g.20gb 1x 1g.10gb |
| mixed-2 | 1x 4g.40gb 3x 1g.10gb |
| mixed-3 | 1x 3g.40gb 2x 2g.20gb |
| mixed-4 | 1x 3g.40gb 1x 2g.20gb 2x 1g.10gb |
| mixed-5 | 1x 3g.40gb 4x 1g.10gb |
| mixed-6 | 3x 2g.20gb 1x 1g.10gb |
| mixed-7 | 2x 2g.20gb 3x 1g.10gb |
| mixed-8 | 1x 2g.20gb 5x 1g.10gb |

### H100 94GB GPU

The following table defines the MIG profiles supported on the H100 94GB NVIDIA
GPU:

| Partitioning Scheme | Available Partitions |
|---|---|
| 1g.12gb | 7x 1g.12gb |
| 1g.24gb | 7x 1g.24gb |
| 2g.24gb | 3x 2g.24gb |
| 3g.47gb | 2x 3g.47gb |
| 4g.47gb | 1x 4g.47gb |
| 7g.94gb | 1x 7g.94gb |
| mixed-1 | 1x 4g.47gb 1x 3g.47gb |
| mixed-2 | 1x 4g.47gb 1x 2g.24gb 1x 1g.12gb |
| mixed-3 | 1x 4g.47gb 3x 1g.12gb |
| mixed-4 | 1x 3g.47gb 2x 2g.24gb |
| mixed-5 | 1x 3g.47gb 1x 2g.24gb 2x 1g.12gb |
| mixed-6 | 1x 3g.47gb 4x 1g.12gb |
| mixed-7 | 3x 2g.24gb 1x 1g.12gb |
| mixed-8 | 2x 2g.24gb 3x 1g.12gb |
| mixed-9 | 1x 2g.24gb 5x 1g.12gb |

### H200 141GB GPU

The following table defines the MIG profiles supported on the H200 141GB NVIDIA
GPU:

| Partitioning Scheme | Available Partitions |
|---|---|
| 1g.18gb | 7x 1g.18gb |
| 1g.35gb | 4x 1g.35gb |
| 2g.35gb | 3x 2g.35gb |
| 3g.71gb | 2x 3g.71gb |
| 4g.71gb | 1x 4g.71gb |
| 7g.141gb | 1x 7g.141gb |
| mixed-1 | 1x 4g.71gb 1x 3g.71gb |
| mixed-2 | 1x 4g.71gb 1x 2g.35gb 1x 1g.18gb |
| mixed-3 | 1x 4g.71gb 3x 1g.18gb |
| mixed-4 | 1x 3g.71gb 2x 2g.35gb |
| mixed-5 | 1x 3g.71gb 1x 2g.35gb 2x 1g.18gb |
| mixed-6 | 1x 3g.71gb 4x 1g.18gb |
| mixed-7 | 3x 2g.35gb 1x 1g.18gb |
| mixed-8 | 2x 2g.35gb 3x 1g.18gb |
| mixed-9 | 1x 2g.35gb 5x 1g.18gb |

### B300 GPU

The following table defines the MIG profiles supported on the B300 NVIDIA GPU:

| Partitioning Scheme | Available Partitions |
|---|---|
| 1g.34gb | 7x 1g.34gb |
| 1g.67gb | 4x 1g.67gb |
| 2g.67gb | 3x 2g.67gb |
| 3g.135gb | 2x 3g.135gb |
| 4g.135gb | 1x 4g.135gb |
| 7g.269gb | 1x 7g.269gb |
| mixed-1 | 1x 4g.135gb 1x 3g.135gb |
| mixed-2 | 1x 4g.135gb 1x 2g.67gb 1x 1g.34gb |
| mixed-3 | 1x 4g.135gb 3x 1g.34gb |
| mixed-4 | 1x 3g.135gb 2x 2g.67gb |
| mixed-5 | 1x 3g.135gb 1x 2g.67gb 2x 1g.34gb |
| mixed-6 | 1x 3g.135gb 4x 1g.34gb |
| mixed-7 | 3x 2g.67gb 1x 1g.34gb |
| mixed-8 | 2x 2g.67gb 3x 1g.34gb |
| mixed-9 | 1x 2g.67gb 5x 1g.34gb |

## What's next

- [Create a shared cluster](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform/pa-user/create-user-cluster)
- [Create a standard cluster](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform/pa-user/create-standard-cluster.md.txt)
- [Maintain Kubernetes clusters](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform-application/pa-ao-operations/cluster.md.txt)
- [Manage node pools](https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/gdcag/platform/pa-user/manage-node-pools.md.txt)