Servers offers three dedicated compute products: Metal, Accelerate, and Build. Each product is optimized for a different type of workload or use case.

Features documented in the Servers section are available for all compute products.


Metal are bare metal instances for general-purpose workloads such as deploying your company applications.


Accelerate are dedicated instances for accelerated computing. They use the newest, most powerful GPUs on the market. Accelerate is recommended for Machine Learning workloads such as inference, fine-tuning, and training.

We only offer dedicated instances, often referred to as bare metal servers. You can find all available instances on the Accelerate pricing page.

Form factor

The g3.large.x86 and g3.xlarge.x86 instances use the PCIe version of the H100 GPU.

GPUs are connected in pairs using NVLink, meaning the g3.large.x86 has 2 NVLink pairs and the g3.xlarge.x86, four pairs.

g3.large.x86 form factor

g3.xlarge.x86 form factor


Accelerate instances are currently available in Dallas and Frankfurt.


Build is a white glove service that allows you to build the infrastructure stack of your dreams without having to worry about owning and maintaining hardware. Build is like having your own global infrastructure team without any of the headaches that comes with it.

We buy, build, install, and maintain the server configurations you choose for your projects in any of our existing locations. We are also able to provision new locations depending on your deployment size.