Bare Metal GPUs for AI/ML Workloads | DigitalOcean

Featured AI Products
Build, deploy, and scale cloud compute resources
Containers and Images
Safely store and manage containers and backups
Fully managed resources running popular database engines
Management and Dev Tools
Control infrastructure and gather insights
Secure and control traffic to apps
Help protect your account and resources with these security features
Store and access any amount of data reliably in the cloud
Security and Networking
Web and App Platforms
Get raw, single-tenant access to the latest NVIDIA H100, H200, and AMD MI300X GPUs. Purpose-built for serious AI teams training foundational models, running large-scale inference, and pushing the boundaries of what's possible.
AMD's Instinct MI300X unleashes cutting-edge compute and memory technologies for best-in-class performance for AI model training, inference, and high-performance computing. Built to handle the most data-intensive AI and high-performance workloads, it ensures seamless data processing, reduces bottlenecks, and enables faster results.
Compare the key specs of Bare Metal GPUs to understand how each can power your AI applications
Talk to an expert to reserve capacity today
What are Bare Metal GPUs?
GPU Memory bandwidth
Monitoring GPU utilization for Deep Learning
Multi-GPU on raw PyTorch with Hugging Face's Accelerate library
How to maximize GPU utilization by finding the right batch size
The Hidden Bottleneck: How GPU Memory Hierarchy Affects Your Computing Experience
Bare Metal GPUs are dedicated, single-tenant servers equipped with eight high-performance GPUs (such as NVIDIA HGX H100, H200, or AMD MI300X) designed to support intensive AI/ML workloads on bare-metal infrastructure.
GPU Droplets are virtualized and great for flexible, on-demand GPU workloads but share infrastructure. Bare Metal GPUs provide isolated hardware without virtualization—offering consistent, high-performance, and customizable environments for demanding AI training, inference, or HPC tasks.
Currently, Bare Metal GPUs are available in New York (USA) and Amsterdam (Netherlands). Additional regions are expected to launch in the future.
Bare Metal GPUs are provisioned via contract with DigitalOcean sales (not on-demand). Billing is handled through Paperspace, requiring a Paperspace account to manage payments.
Yes—you have root-level control. Servers come preconfigured with recommended OS (Ubuntu 22.04 or 24.04), GPU drivers, CUDA, or ROCm—but you can modify or install any software, driver, or toolkit you need.
Bare Metal GPUs have no default firewall. It's recommended to use UFW or your preferred firewall to secure the public interface. IPMI access is not provided.
© 2026 DigitalOcean, LLC.