Skip to main content

Create a VM with a GPU

This guide creates an Ubuntu VM with an NVIDIA GPU from the Hikube console, then checks that the GPU is usable. For GPUs in a Kubernetes cluster, see Provision a GPU on Kubernetes.

Prerequisites​

  • A Hikube account and a project (see Hikube quick start).
  • Available quotas: at least 8 vCPU, 32 GB of memory and 50 GB of storage.
  • A public SSH key (cat ~/.ssh/id_ed25519.pub).

Step 1: Open the creation wizard​

  1. In the side menu, open Infrastructure > VM Instances.
  2. Click Create an Instance.
  3. General step: enter the Instance name, for example vm-gpu01, then Next.

Step 2: Configure and validate​

Configuration: size and GPU​

  1. Under Resources (CPU / RAM), choose Universal (U) > 2XLarge (8 vCPU, 32 GB).
  2. Under Hardware Acceleration (GPU), click the NVIDIA L40S card (or another available model). The 1 GPU total badge appears. The card's + and − buttons adjust the number of GPUs.
  3. Click Next.

Models marked Unavailable cannot be selected for the moment.

Storage​

  1. Under Operating System, select ubuntu, version 24.04.
  2. Set Size (GB) to 50: the drivers, CUDA and the ML frameworks take up several tens of GB.
  3. Click Next.

Network​

  1. Leave Public IPv4 Address enabled and SSH (22) checked in Allowed Ports.
  2. Add your key in Authorized SSH keys.
  3. Optional: enable Cloud-Init script (User Data) to install the drivers automatically (script in Install CUDA).
  4. Click Next.

Summary​

The Summary shows a Hardware Acceleration (GPU) line with the model and quantity. Check the estimated cost, then click Create instance.

Step 3: Check the status​

In the VM Instances list, wait for the Running status. On the detail page, the Resources & Characteristics section lists the GPU under GPUs.

Expected result: Running status and one badge per GPU under GPUs, in Resources & Characteristics. The badge shows the technical name of the model (for example l40s for an NVIDIA L40S).

Step 4: Retrieve the connection details​

Copy the command from the SSH Connection block (Network & Security section of the detail page), for example ssh ubuntu@203.0.113.20.

Step 5: Connect and test​

ssh -i ~/.ssh/id_ed25519 ubuntu@203.0.113.20

# The GPU is visible on the PCI bus
lspci | grep -i nvidia

Expected result:

06:00.0 3D controller: NVIDIA Corporation ...

Then install the NVIDIA drivers and CUDA by following Install CUDA and GPU drivers, then:

nvidia-smi

Expected result: the nvidia-smi table lists the GPU (for example NVIDIA L40S) with its memory.

Step 6: Quick troubleshooting​

SymptomAction
The Hardware Acceleration (GPU) section does not appearThe platform offers no GPU for the moment: contact support.
All cards are UnavailableNo free GPU: try again later or contact support.
The following GPUs are not available: … on creationThe requested GPUs are not free together on the same server: reduce the number of GPUs or change model.
lspci shows no NVIDIA GPUCheck on the detail page that the GPU is listed; otherwise, add it via Edit.
nvidia-smi: command not foundThe drivers are not installed: see Install CUDA.

More cases in the GPU troubleshooting page.

Step 7: Cleanup​

  1. On the VM's detail page, click Delete.
  2. Enter the VM's name and click Permanently delete.
  3. The system disk remains in the Disks menu: delete it from that menu if you no longer need it.
Stop rather than delete?

Stopping the VM (Stop) releases the GPU, which can be assigned to another workload: on restart, you may have to choose another model. Delete the VM if you no longer need it, stop it if you intend to start it again.

Next steps​