Docs

Volcengine Kubernetes Engine

Creating clusters

Creating a managed heterogeneous computing cluster

VKE supports quick creation of managed clusters for heterogeneous computing in AI inference scenarios. This topic describes the procedure for creating a managed cluster for heterogeneous computing.

Cluster description

A managed cluster for heterogeneous computing involves simplified cluster parameter configuration, which enables quick cluster creation for heterogeneous computing scenarios. A managed cluster for heterogeneous computing offers the same capabilities as a standard managed cluster.

Prerequisites

You have already registered an account on the Volcano Engine platform and completed necessary preparations such as identity verification, account recharge, and authorization. For details, see Preparations.

Your Volcano Engine account balance is not zero when coupons are excluded and not less than CNY 100 when coupons are included. This prerequisite ensures that you can successfully create pay-as-you-go resources. For details, see Billing center.

Permissions to use Volcengine Kubernetes Engine (VKE) and for VKE to access cloud resources have been granted. For details, see Cloud service dependencies.

‌Check the quota limits for the VKE cluster and the dependent cloud services. For details, see Limits.

Currently, some GPU Accelerated and High Performance Computing Optimized instance specifications, for example, ecs.hpcpni3ln required for deploying DeepSeek-R1, are in the beta stage. If you want to use a GPU Accelerated or High Performance Computing Optimized instance specification in the beta stage, ensure that you have requested it and understood the regions, availability zones, and billing methods of the instance specification.

To use the High Performance Computing Optimized instance specification to create a managed heterogeneous computing cluster, you can create an HPC cluster for nodes in advance. For details, see Creating HPC clusters.

tip

The HPC cluster must be deployed in the same availability zone as the VKE node.

A VMP workspace has been created. For details, see Creating workspaces.

Cost information

Creating a heterogeneous computing cluster may incur costs for cloud infrastructure resources such as load balancers, NAT Gateway instances, and Elastic IP addresses. You can see actual billing information by checking the Configuration fee field in the right side of the Create cluster page. For billing details, see Product billing.

Procedure

Select the target region in the top navigation bar.

In the left navigation bar, click Cluster. On the page that appears, click Create cluster.

In the Create cluster dialog box, go to the Managed clusters for heterogeneous computing section and click Create now.

Configure cluster parameters described in the following table. For parameter details, see Creating managed clusters.

Parameter	Description
Cluster and node pool configuration
Availability zone	Select the availability zone to which the node pools in the managed heterogeneous computing cluster belong. warning To use the High Performance Computing Optimized specification, the availability zone that you select needs to include the availability zone where the HPC cluster resides.
Node-based billing model	The billing method for the ECS resources within the node pool. You can select one of the following options: Pay-as-you-go: a postpaid mode in which you use first and pay later Subscription: a prepaid model in which payment is made in advance before use. tip For ECS billing details, see Product billing.
Instance type	Managed heterogeneous clusters support High Performance Computing Optimized and GPU Accelerated instance types. Select an instance type and specification that best suits your business scenario. For a detailed description of the specifications, see Instance types and specifications.
High-performance computing cluster	This parameter is available if you select High Performance Computing Optimized for Instance type. Select the HPC cluster to which the HPC optimized nodes belong and its project. If no high-performance compute cluster is available, click Create high-performance computing cluster to create one. For details, see Creating HPC clusters. warning The availability zone for the HPC cluster must be the same as the availability zone selected for the VKE cluster.
Number of nodes	Enter the number of nodes to create. The total number of nodes, including new and existing ones, cannot exceed the cluster's node quota. For quota details, see Limits. Warning The expected number of nodes on the VKE side and the ECS side varies. For details, see FAQ.
Login credentials	Select a node login method. Password: Enter the root password twice to confirm. SSH key pair: You use an SSH key pair (a public key and a private key) to log in to the node. This login method is more secure and available to Linux instances. tip You must configure a node login method, regardless of whether the node is a newly created or existing one. This is because each node in the created cluster will have a new operating system. The login method specified here will also apply to other nodes in the node pool where this node resides.
Networking
VPC	Select the VPC to which the managed heterogeneous computing cluster belongs. A VPC provides an isolated, self-configurable, and self-managed virtual network environment for the cluster. If no VPC is available, follow the steps described in Creating VPC to create one.
Pods and Node Subnets	Select the subnet to which the pods and nodes of the managed cluster for heterogeneous computing belong. If no subnet is available, click Create subnet to create one. For detailed steps, see Creating subnets.
Observability configuration
Container observability	Configure basic container monitoring and control-plane service monitoring to monitor metrics of fundamental resources (nodes, kubelet, kube-state-metrics, and more) and control-plane services (kube-apiserver, kube-scheduler, etcd, and more). Configure the observability metric scope. Available options include: Basic: (Default) Basic metrics for the monitored object. They provide insights into the basic status of the object and resource usage. Standard: (Recommended) Basic metrics plus important metrics of the monitored object. The combination of basic and important metrics provides a more comprehensive view of the object status. Full set of metrics: Basic, important, and advanced metrics of the monitored object. This tier includes all metrics of the monitored object. tip The billing method and cost vary based on the observability metric scope and the VMP workspace's storage duration. For details, see Pay-as-you-go. Prometheus workspace: Select a VMP Prometheus workspace where VKE observability metrics are stored. tip This configuration is only used to bind the workspace to the cluster. It will not collect metric data. To collect data, install the prometheus-agent add-on in the cluster. For details, see Configuring the add-on or Installing the add-on. If no VMP workspace is available, create one first. For details, see Creating workspaces. Observational enhancement: If you enable this feature, when any of the core add-ons (core-dns, vpc-cni, and csi-ebs) are installed in the cluster, the system will automatically enable observability for these add-ons and provide dashboards and alerting capabilities. For details, see DNS service observability, CNI network observability, and CSI storage observability.
Advanced settings
Control plane subnets	Select a subnet that is used for the Kubernetes API server to create a load balancer. If you do not specify a subnet, the system will randomly select one. If no subnet is available, click Create subnet to create one. For detailed steps, see Creating subnets.
Data volume	Additional storage space for daily operational data. A single node can have up to 15 data volumes. Warning It is recommended to keep the system default data volumes in a managed cluster for heterogeneous computing. The date01 directory will be used as a dependency later and therefore cannot be modified. Ensure the data that needs long-term storage has been backed up properly in case of computer crashes or disk damage which causes data loss.

Confirm the configuration details and configuration fee. If nothing is wrong, select the checkbox before the Terms of Service and click Create now.

Subsequent operations

After the managed heterogeneous computing cluster is created successfully, you can perform heterogeneous computing, AI inference, and service deployment. For detailed instructions, see AI cloud native.

Last updated: 2026.07.24 10:53:32