How to run GPU workloads on B200 or B300 GPU on Verda

Let's discover how to run GPU workloads on Verda (formerly DataCrunch).

Getting Started

Verda is a Finland-based hyperscaler with access to modern NVIDIA GPUs like B300, B200, H200, H100, and RTX Pro 6000. It is a good choice when you need high-end GPU capacity with flexible deployment options.

Get started by creating an account at https://console.verda.com. For first use, you may need to top up your account (for example, minimum $20).

What you'll need

• Verda account
• SSH public key
• Basic Linux/SSH familiarity

Launch a GPU instance

1. Create a new instance from the Verda console.

2. Choose pricing mode:

   • Spot — lower cost, but can be interrupted
   • On-demand — stable capacity, higher cost

3. Select the GPU type you need (for example: B300, B200, H200, H100, A100, RTX Pro 6000, or L4).

4. Configure node details:

   • SSH key
   • OS image (for example, Ubuntu 24.04 + CUDA)
   • Storage and networking

5. Launch the instance and wait until it is in the running state.

6. Connect over SSH and verify GPU access:

nvidia-smi

Cost optimization tips

• Use spot instances for non-critical jobs and experiments.
• Use on-demand for production or long-running workloads requiring stability.
• Pick the right GPU tier for your use case; avoid over-provisioning.
• Shut down idle instances to control hourly spend.

Screenshots

Here are some screenshots to help you visualize the process.

Verda Console Dashboard

Instance Configuration

Running instance

Running nvidia-smi command on instance

Sample docker benchmark details