Skip to main content

Creating a Kubernetes cluster

Overview​

Setting up a managed Kubernetes cluster on Civo offers a smooth experience, enabling you to personalise different components of your cluster for an ideal configuration. Civo also provides a number of CPU and GPU options to ensure optimal performance, cost-efficiency, and scalability.

If you run GPU workloads (e.g. TensorFlow, PyTorch, Real-time Inference, etc.) on Kubernetes, you will need higher computational requirements. Large models will require nodes with more GPU memory.

Deploy Standard, Performance, or CPU-Optimized Workloads on Civo Kubernetes​

The following instructions outline the steps for creating a Kubernetes cluster for standard workloads, with options to set up the cluster using the Civo Dashboard, Civo CLI, or Terraform.

Creating a cluster on the Dashboard​

Begin by selecting the Civo Region you are operating in. You can do so in the lower left of your Dashboard page:

Region selection menu

Then, navigate to the Kubernetes cluster creation page:

Overview of the cluster creation page on Civo

The numbered sections give you options for the specifications of your cluster.

1. Name​

This is your cluster's name. May not contain a space.

2. Number of nodes​

Kubernetes clusters on Civo can have multiple node pools - this is the number you want to launch in the cluster's initial node pool. This can be scaled up and down once the cluster is running, and does not include the control plane node, which is handled by Civo.

3. Network​

If you have created custom networks in this region, you will be able to choose one of them here. Custom networks allow you to define private networking within a region, which allows you to prevent some resources from being routable to the public internet. Civo Kubernetes clusters will always have a public IP address.

4. Firewall​

If you have configured existing firewalls in this region, you will be able to select one of them for this cluster, or create a new firewall and specify which port(s) to open. Firewall rules can be customised after creation.

5. Node size​

The specifications for the machines in the initial node pool. These are priced per node, per hour. The "Hourly/Monthly" slider allows you to estimate the cost per month or per hour of the cluster running. For more information, see the Billing section.

Sizes or configurations may not be available due to quota on your account or the number of nodes you have chosen in section 2 above.

Depending on the applications you want to run on your cluster, you may need to select larger nodes.

Advanced options and marketplace

6. Advanced options​

This section allows you to optionally configure advanced options.

Container Networking Interface (CNI)​

The default CNI on Civo is Flannel. However, you can choose Cilium as an alternative Container Networking Interface (CNI) for your cluster.

Please note that the Cilium CNI is not compatible with Talos clusters. If you choose Cilium as your CNI and set Talos as your cluster type, the system will default to using Flannel.

Cluster type​

The Cluster type selector allows you to choose between K3s or Talos Linux. The underlying operating system on K3s clusters is an Alpine Linux image. Talos Linux is an immutable Kubernetes-oriented Linux operating system.

As noted above, please note that the Cilium CNI is not compatible with Talos clusters and Flannel will be used instead.

7. Marketplace​

You can remove applications to prevent them from being installed by default, or add applications to start alongside your cluster. For more information, see the Marketplace documentation.

Creating your cluster​

When you are satisfied with your initial cluster configuration, you can click "Create cluster" and be directed to the cluster's dashboard page. It will take a moment to become active, and you will be shown the status throughout.

Cluster building

Once running, you can use kubectl and the downloaded kubeconfig file from the cluster's page to interact with your cluster. You will find the kubeconfig file for download in the "Cluster information" section:

Cluster information showing Kubeconfig download

Deploy GPU Workloads on Civo Kubernetes​

Running GPU workloads on Kubernetes is becoming increasingly common due to the flexibility and scalability it provides. Deciding on the type of Kubernetes nodes for your GPU workloads involves many considerations. Whether you require low-latency responses, high-throughput inference requests, or other performance needs, Civo has you covered.

Civo provides the following GPU Types:

  • NVIDIA A100 Tensor Core GPU: Available in both 40GB and 80GB variants, this GPU is designed for high-demand workloads such as machine learning model training, large language models, and scientific computing. It offers significant computational power with over 312 teraflops of FP16 performance and 1,248 Tensor cores.
  • NVIDIA H100 Tensor Core GPU: Known for its advanced Hopper architecture, this GPU excels in AI training and inference tasks, making it ideal for developing and deploying large AI models like chatbots and recommendation engines.
  • NVIDIA L40S GPU: With 48GB of GDDR6 memory, this GPU is suitable for tasks requiring a blend of AI computations and advanced graphics processing, such as 3D graphics rendering and training large language models.
  • NVIDIA GH200 Grace Hopper Superchip: This GPU features an integrated CPU-GPU architecture tailored for generative AI, large-scale AI inference, and high-performance computing workloads that demand substantial memory and processing power.

To deploy GPU workloads on Kubernetes, add a new node pool to your cluster and install the Kubernetes operator for GPU nodes

note

GPU nodes for Kubernetes is a separate node SKU than traditional Kubernetes nodes.