Overview

This section of the documentation covers our previous generation of Gradient. For the current version go to Gradient Next.

Gradient clusters are private clusters that run machine learning workloads. Gradient clusters can be created on Paperspace Cloud, on any other cloud provider (AWS, GCP, Azure), or on your own servers via the Gradient Installer.

Find out more about Gradient's multi-cloud capabilities here.

You can create a managed cluster using the Web UI in a couple of clicks.

Choosing between our managed service, managed private clusters, and self-hosting Gradient

	Managed Service (multi-tenant)	Managed Private Clusters	Self-Hosted Clusters
Infrastructure:	Shared, managed by Paperspace	Private, managed by Paperspace	Private, self-hosted
Setup time:	None	Setup time: 10 minutes	Setup time: 20-30 minutes
Features:	Notebooks (including Free GPUs!), basic experiments	Notebooks, experiments, deployments, model repo, data management, GradientCI	Notebooks, experiments, deployments, model repo, data management, GradientCI
Target audience:	Hobbyists & students	Startups & SMBs running production workloads	Mid-market & enterprise businesses conducting ML at scale

This section of our documentation covers the private cluster options. If you are looking to use our managed service, just create an account to get started right away.

Cluster pricing

Managed

Compute, Storage, & Networking The Kubernetes master node, storage, and networking cost to run the cluster is $0.12/hr. Private Clusters require a minimum of one CPU node for cluster orchestration and clusters include 500GB of storage by default.

In addition, instances used to run workloads are charged at the regular rate (see instance pricing) plus a small compute premium.

Self-Hosted

Compute, Storage, & Networking Customers are responsible for their infrastructure costs. Gradient does not bill for any compute, storage, and networking costs other than the compute premium.

Subscription Gradient Private Clusters require a Essentials or great subscription.