Gozunga Cloud offers GPU-accelerated instances powered by NVIDIA L40S and RTX 2000 PRO (Blackwell) hardware. With a single cloud-init script, your instance boots with NVIDIA drivers and CUDA installed — no manual setup required.

1. Overview
2. Available GPU Instance Types
3. Create Your Gozunga Account
4. Launch a GPU Instance via the Portal
5. Launch via the OpenStack CLI
6. Connect and Verify

1. Overview

Gozunga Cloud GPU instances give you bare-metal GPU performance in a flexible cloud VM. Whether you're running AI inference, training models, rendering, or scientific computing, you can be up and running in minutes.

We use cloud-init to handle driver installation automatically at boot time. The script blacklists the open-source Nouveau driver, installs the official NVIDIA server drivers (open-kernel variant) and nvtop, then reboots — leaving your instance ready to use in approximately 5 minutes from launch.

The cloud-init configuration is open-source and available in the Gozunga cloud-init collection on GitHub.

2. Available GPU Instance Types

Both instance types run on Ubuntu 26.04 and are compatible with the same cloud-init configuration.

💡 Pro Tip: The RTX 2000 PRO (Blackwell) is Gozunga's latest GPU offering — built on NVIDIA's Blackwell architecture for exceptional inference performance per watt.

3. Create Your Gozunga Account

If you haven't yet, Create an Account. New accounts receive $100 in free credits — enough to get meaningful work done before you commit to a plan.

4. Launch a GPU Instance via the Portal

The easiest way to get started is through the Gozunga Cloud Management Portal.

1. Log in and create or select a Project.
2. From the Servers pane, click Create a Cloud Server.
3. Select Ubuntu 26.04 as your image.
4. Under hardware type, select Accelerated Compute, then choose your GPU instance type — ac.nvidia_l40s1 (L40S) or ac.nvidia_rtx2000pro1 (RTX 2000 PRO).
5. Select your network (e.g. Internet) and your SSH key.
6. Under Security Groups, add SSH to allow inbound access.
7. Under Cloud Configuration, paste the cloud-init YAML below:

#cloud-config
autoinstall:
  version: 1

# Update/upgrade the system first
package_update: true
package_upgrade: true

# Blacklist the Nouveau driver (required before NVIDIA driver install)
write_files:
  - path: /etc/modprobe.d/blacklist-nouveau.conf
    content: |
      blacklist nouveau
      blacklist lbm-nouveau
      options nouveau modeset=0
      alias nouveau off
      alias lbm-nouveau off
    permissions: '0644'

runcmd:
  - update-initramfs -u
  - add-apt-repository ppa:graphics-drivers/ppa -y
  - apt update -y
  - apt install -y nvidia-driver-595-server-open nvtop

power_state:
  mode: reboot
  message: "Rebooting to load NVIDIA drivers"
  timeout: 1

8. Click Create and wait for the instance to reach Active status.

The instance will boot, run the cloud-init script, and automatically reboot to load the NVIDIA drivers. Allow approximately 5 minutes from launch before connecting.

5. Launch via the OpenStack CLI

If you prefer the command line, you can use the OpenStack CLI directly. First, download your OpenStack RC file from the portal and source it:

source your-project-openrc.sh

Save the cloud-init script locally (or download it directly from GitHub):

curl -o ubuntu26-with-cuda-and-opensource-nvidia-drivers.yaml \
  https://raw.githubusercontent.com/gozunga/cloud-init-collection/main/ubuntu26-with-cuda-and-opensource-nvidia-drivers.yaml

Launch an RTX 2000 PRO (Blackwell) instance:

openstack server create \
  --image ubuntu-26.04 \
  --flavor ac.nvidia_rtx2000pro1 \
  --key-name key-file \
  --network Internet \
  --security-group SSH \
  --user-data ./ubuntu26-with-cuda-and-opensource-nvidia-drivers.yaml \
  --wait \
  rtx-vm

Or launch an L40S instance:

openstack server create \
  --image ubuntu-26.04 \
  --flavor ac.nvidia_l40s1 \
  --key-name key-file \
  --network Internet \
  --security-group SSH \
  --user-data ./ubuntu26-with-cuda-and-opensource-nvidia-drivers.yaml \
  --wait \
  l40s-vm

The --wait flag holds until the instance reaches Active status. Once it does, wait another ~4 minutes for cloud-init and the automatic reboot to complete before connecting.

6. Connect and Verify

Once the instance has had time to complete its setup (~5 minutes after Active), connect via SSH:

ssh ubuntu@<your-instance-ip>

Verify the NVIDIA driver loaded correctly:

nvidia-smi

You should see output similar to:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.x.x       Driver Version: 595.x.x     CUDA Version: 12.x               |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|=========================================+========================+======================|
|   0  NVIDIA RTX 2000 PRO             Off|   00000000:00:08.0 Off |                  Off |
|  N/A   35C    P0              30W / 200W |       0MiB / 16384MiB |      0%      Default |
+-----------------------------------------------------------------------------------------+

You can also use nvtop for a real-time GPU activity monitor:

nvtop

Your GPU instance is ready. Install your frameworks (PyTorch, TensorFlow, JAX, etc.) and start building.

Need help? Our support team is available 24/7. Open a ticket from the Gozunga Portal or use the support widget on this page.