Skip to main content

Fluidstack Atlas

Fluidstack Atlas is an AI IaaS platform designed from the ground up to meet the infrastructure requirements of top AI labs. Atlas has extremely flexible deployment requirements, and can be installed on any NVIDIA HGX Reference Architecture cluster.

Deployment Models​

Atlas supports three primary deployment models.

Single Tenant Slurm​

A single tenant Slurm region is composed of a single, large, slurm cluster managed by the Atlas control plane. This deployment variant offers the best performance, scale, and feature set for large training clusters.

Single Tenant Kubernetes​

A single tenant Kubernetes region is composed of a single, large Kubernetes cluster managed by the Atlas control plane. This deployment variant offers the best performance, scale, and feature set for large inference clusters.

Multi Tenant Cloud​

A multi-tenant cloud region is a fully-featured cloud region built on top of a highly-tuned kvm virtualization stack that can be be deployed on any bare metal cluster that follows the NVIDIA HGX reference architecture.

Partnerships​

Atlas can be quickly deployed on most standard clusters. Whether you have an existing bare metal cloud platform or are looking to monetize a private training cluster, Atlas provides the flexibility to adapt to your infrastructure. Reach out to us at sales@fluidstack.io to learn more about deployment options.