How to Create Bare Metal GPUs

The process of purchasing and provisioning DigitalOcean Bare Metal GPUs has the following steps:

1. Complete a contract for the bare metal GPUs you want.

   Bare metal GPUs are not available for on-demand provisioning, so first, contact our sales team to finalize the price and provisioning.

2. Set up billing with your Paperspace account.

   Payment for bare metal GPUs is managed through Paperspace, a platform to manage GPU-based applications on DigitalOcean. To handle billing for bare metal GPUs, create a Paperspace account if you don’t already have one.

3. Connect to your servers with the provided IP addresses.

   Once you complete the contract and we provision your resources, we email you the IP addresses of your servers, and you can begin using them.

Complete a Contract

When you contact sales, you choose the hardware and software configuration you want. You can view details of the available hardware and software options on the features page:

Bare Metal GPU Features

Learn more about bare metal GPU features and view comparisons to related products to help you choose the right product for your needs.

All software configurations come with the necessary NVIDIA GPU and CUDA drivers.

Set up Billing

After you complete a contract with sales, you create a Paperspace account to manage billing.

Connect to Your Server

Once we provision your bare metal GPUs, we email you the IP address of the servers, and you have full control of the configuration. You can connect to it using SSH.

Once you connect, you may want to:

• Set up UFW to protect your bare metal GPU on its public interface. By default, there are no firewalls or gateways in front of bare metal GPUs.

• Install Miniconda, a minimal installation of Anaconda, an AI-focused operating system. Miniconda includes Python, the conda package manager, and a limited number of additional packages and dependencies.

• Install Transformers, an ML toolkit and collection of pre-trained models.

• Install PyTorch, a Python ML library that has support for tensor computing with GPU acceleration and deep neural networks.

If you have any issues, reach out to support.