API How-Tos

The IONOS CLOUD API lets you manage Cloud GPU VMs programmatically using conventional HTTP requests. You can use the API to create, delete, and retrieve information about your Cloud GPU VMs.

Furthermore, you need templates to provision Cloud GPU VMs, but templates are not compatible with servers that support full flex configuration.

Cloud GPU VM workflow

To get started with Cloud GPU, follow these steps:

1

Verify service readiness

Cloud GPU VM activates automatically so you can begin provisioning immediately. We recommend contacting the IONOS CLOUD Support to verify that your contract is ready and to ensure your resource limits meet your project requirements.

2

Discovery and selection

Cloud GPU VMs are available in preconfigured sizes. Select a template that fits best to your needs and matches your contract's resource limits, which will be adjusted as part of the access enablement process. It is not possible to change the size of a Cloud GPU VM instance once it has been provisioned.

Resource limits: By default, you can deploy only 1 Cloud GPU VM using the H200–S template. To deploy templates sized M, L, or XL, or to run multiple S instances, you must first request a resource limit increase through the IONOS Cloud Support.

• List all Cloud GPU VM templates.

• Review Cloud GPU VM template specifications to select the appropriate template for your needs.

• Use the template UUID to make an API request call for creating a new Cloud GPU VM.

3

Create a Cloud GPU VM

Initiate a Cloud GPU VM creation request through API with the following:

• The selected GPU template UUID

• Linux-based operating system

Note: The product supports only Linux-based operating systems during the launch.

• Network settings

• SSH keys, Cloud-init script, or other authentication methods

Result: The VM with the attached GPU is provisioned.

4

Access and setup

• Connect to VM using one of the following options:

  • Through the SSH.

  • Through the remote console.

• Install NVIDIA drivers. For more information, refer to the NVIDIA documentation.

• Run the following command in your terminal or command prompt to verify if the Cloud GPU VM is accessible: nvidia-smi

Result: A successful output should display the driver version, the GPU model(s), the GPU temperature, and the current memory usage. If the command is not found or returns an error, your drivers may be missing or the GPU is not properly installed.

• Install required framework dependencies. Use a package manager like pip or conda to install your chosen framework, ensuring you select the GPU-enabled version.

For PyTorch

Install the version that matches your CUDA toolkit. For example, CUDA 12.1.

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

For TensorFlow

• Use the server for your use case

5

Usage

• Run GPU workloads, such as model training (finetuning), inference, or graphics rendering

• Monitor GPU utilization and performance metrics

6

Management

• Start, restart, or delete the Cloud GPU VM as needed.

• Monitor costs and usage. For more information, see Cost Alert and Cost & Usage.

7

Cleanup

• Back up your results and data before your delete your VM. For more information, see Install Acronis Backup Agent on Linux.

• Delete the Cloud GPU VM from the VDC.

Quick Links

Retrieve Template list

Learn how to retrieve a list of available templates.

Retrieve Template details

Learn how to retrieve details of a specific template.

Create a Cloud GPU VM

Learn how to create a Cloud GPU VM.

Delete a Cloud GPU VM

Learn how to delete a Cloud GPU VM.