Skip to main content

Create a cluster

Creating a cluster gets you GPU-backed worker nodes running on demand, without provisioning dedicated bare metal capacity yourself. In the Hanoi 2 and Japan regions, this service supports the NVIDIA H100 SXM5 and H200 SXM5 GPU cards.

Requirements​

Complete Initial setup first, then confirm:

  • CPU, GPU, RAM, Storage, and Instance quota sufficient for your cluster configuration. If you use Autoscaling, quota must also cover the worker group's configured maximum node count.
  • A network subnet with a Static IP Pool, for the Kubernetes nodes.

Create the cluster​

  1. On the FPT Cloud Portal menu, select Containers > Kubernetes > Create a Kubernetes Engine.

  2. Enter the cluster's basic information, then select Next:

    • Name: a name for the cluster.
    • Network: the subnet to deploy the cluster's VMs on.
    • Version: the Kubernetes version.
    • Cluster Endpoint Access: the access option for the cluster's Kubernetes endpoint. Choose based on your security requirements and network architecture. If you select Public & Private or Private, an Allow CIDR field appears — enter the IP ranges allowed to reach the endpoint.

  3. Configure node pools, then select Next.

    The portal doesn't support creating GPU workers as the base worker group — create GPU workers starting from worker group 2 onward.

    Base worker group (general-purpose, non-GPU):

    • Instance Type: General.
    • Type: the CPU/Memory configuration for the worker nodes.
    • Container Runtime: Containerd.
    • Policy: the Storage Policy (IOPS tier) for the worker node disk.
    • Disk: the root disk size for the worker nodes.
    • Scale min / Scale max: the worker group's minimum and maximum node count. We recommend a minimum of 3 nodes for production.
    • Label: a label for the worker group.

    Worker group 2 and beyond (GPU):

    • Instance Type: GPU.
    • GPU type: NVIDIA H100 SXM5 or NVIDIA H200 SXM5.
    • GPU sharing configuration.
    • GPU configuration (CPU/RAM/GPU RAM).
    • Label: a label for the worker group.
    note

    GPU Driver Installation Type offers two options: Pre-install, which adds the NVIDIA GPU driver to the cluster automatically, and User-install, which lets you pick the driver version yourself. A driver lets the worker node's OS (Windows, Ubuntu, and similar) communicate with the GPU — without one, the OS can't use it.

  4. Select Create and review the cluster's initialization information.

  5. Monitor the cluster's creation status. Once it shows Successed (Running), the cluster is ready to use.