Vultr DocsLatest Articles

Vultr Cloud GPU Quickstart

Vultr Cloudnvidia-smiCloud GPU
Updated on July 25, 2024
Vultr Cloud GPU Quickstart header image

This guide explains how to deploy a Vultr Cloud GPU server for tasks such as deep learning, data analytics, and High-Performance Computing. You can use Cloud GPU instances to match your workloads to the processing power required, saving you time and expense. By following this guide, you'll have a fully-functional Cloud GPU instance with all the drivers and libraries you need to get started.

Getting Started

Deploying a Cloud GPU instance is as simple as deploying any Vultr cloud server. Vultr offers two types of GPU servers:

  • Cloud GPU with either full or fractional GPU cards
  • Bare Metal servers with optional GPU cards

Deploy a Cloud GPU

  1. Select Cloud GPU or Bare Metal with Optional GPU as the server type.

    Choose Server

  2. Choose your location.

  3. Choose your distribution, or choose a Cloud GPU Marketplace App.

  4. Choose the amount of GPU memory needed for your application.

  5. Choose your server options, such as automatic backups, IPv6, DDOS protection, and SSH keys.

  6. Give your server a hostname and a label to identify it in the customer portal, then click Deploy Now.

The entire deployment and software installation takes several minutes. During the first boot, cloud-init installs the NVIDIA drivers, CUDA Toolkit, and the CUDA Deep Neural Network (cuDNN) library. You may see the installation in progress if you connect to the Web Console soon after the server deploys.

cloud-init-nvidia-install

After these steps, the server will reboot.

Wait until the final reboot is complete before you begin using the server.

Next Steps

Click your server in the Customer Portal to view the Server Details page. Use the IP address and account information to connect to your server as the root user.

Run nvidia-smi to verify that the NVIDIA drivers and CUDA Toolkit are installed successfully. This command should output details about the driver version, CUDA version, and the GPU itself. For example:

root@TEST-GPU-SERVER:~# nvidia-smi
Thu May 01 16:20:00 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.47.03    Driver Version: 510.47.03    CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  GRID A100D-4C       On   | 00000000:04:00.0 Off |                    0 |
| N/A   N/A    P0    N/A /  N/A |      0MiB /  4096MiB |      0%      Default |
|                               |                      |             Disabled |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

Cloud GPU Marketplace Apps

We have several Cloud GPU Marketplace apps, which are pre-loaded environments for GPU development. You'll find these on the Marketplace Apps tab of the deployment page. Select the GPUC category to browse the apps.

Cloud GPU Marketplace

You'll find more details about each of our apps in the GPU Compute Marketplace.

More Information

See these resources to learn more.