Create GPU Instance

After creating a GPU instance, you can run deep learning training, model fine-tuning, inference services, or other GPU-accelerated tasks on the instance. GPU instances use bare-metal resources. When creating one, you must configure the billing mode, region, network, specifications, image, storage, logon credential, and optional components.

​

Procedure

1. Log in to the GPU Cloud console and go to the Instance page.
2. On the GPU Instance tab, click Create GPU Instance.
3. In the Billing mode section, select a billing mode.

   Parameter	Description
   Pay-as-you-go	Billed based on actual usage duration. Suitable for short-term tasks, temporary validation, or tasks with an uncertain runtime.
   Subscription	Billed by subscription period. Suitable for tasks that run stably over the long term. Subscription durations of 1 month, 3 months, 6 months, 12 months, and 24 months are available.

4. In the Basic Information section, configure the HostName and Region.

   Parameter	Description
   HostName	The instance hostname. You can select IP Address-based Hostname and specify a hostname prefix, or select Custom and enter a hostname. When creating instances in batches, running multi-node tasks, or distinguishing node order by hostname, after selecting Custom, you can optionally select Custom Hostname Order to make hostnames easier to identify and manage later.
   Region	The region where the instance is located. The region cannot be changed after the instance is created.

5. In the Instance & Image section, select the network, specifications, Image, and runtime environment.

   Parameter	Description
   Instance	GPU Instance currently uses a Bare Metal Instance.
   RDMA	Select the RDMA network to be used by the instance.
   Instance Spec	Select a specification that meets task requirements. Specification cards display vCPU, Memory, GPU model, Price, and the number of instances that can be created.
   Image	Select the base OS image to be used by the instance.
   Install GPU Driver	Choose whether to install the GPU driver. When running GPU tasks, it is recommended that you confirm the driver version based on the image, CUDA, framework, and task requirements.
   Driver Version	If Install GPU Driver is selected, select the GPU Driver Version.
   Container Network Mode	Used to configure the network initialization mode for the instance.

   If your workload has no specific container network requirements, we recommend keeping the default options. If existing workloads depend on legacy provisioning network behavior, select Use Legacy Provisioning Behavior.
6. In the Storage section, configure the data disk.

   Parameter	Description
   Mount Data Disk	Select whether to mount a data disk.
   FsType	Select the file system type for the disk. Currently, ext4 and xfs are available.

7. In the Authentication section, configure the Logon Credential.

   Parameter	Description
   Logon Credential	Select Custom Password or Key Pair.
   Logon Username	Select a logon username. Currently, root and ces-user are supported.
   Logon Password / Confirm Password	When Custom Password is selected, enter and confirm the logon password.
   Select Key Pairs	When Key Pair is selected, choose an existing key pair. If you plan to use key pair logon but do not currently have an available key pair, create a key pair first.

8. In the Options section, select whether to install optional components.

   Parameter	Description
   Install Assistant	After installation, the platform can assist in maintaining CES instances.
   Install Monitor	After installation, instance monitoring data can be collected and used for monitoring and alerts.

9. Check the Estimated Cost on the right, and confirm the Instance Type, quantity, and Balance information.
10. Click Continue to proceed to the Review step.
11. After verifying that the configuration is correct, submit the creation request. Once creation is complete, confirm that the instance status is as expected on the Instance page.