Managed clusters · Managed Kubernetes
Kubernetes clusters, without running Kubernetes.
Create a cluster from the console, pick its node pools and add-ons, and deploy your containers. Supergate runs the control plane and keeps the cluster healthy.
One cluster, every kind of workload.
Training jobs, inference services and notebooks share a cluster and its storage, each on the nodes it needs.
Training job
- k8s
- GPU × 8
Inference service
- vLLM
- GPU × 1
Notebook
- JupyterLab
Shared storage
- 2 TB · RWX
What you get.
A managed control plane
Create, inspect, scale and delete clusters from the console. Supergate runs and upgrades the control plane.
Node pools for each workload
Give training, serving and notebooks their own node pools, and resize a pool as the work changes. [TBD: autoscaling]
Metrics and logs built in
See metrics, logs and health signals for each cluster and the workloads on it, in the same place.
Kubernetes and Slurm in one console.
Teams that train on Slurm and serve on Kubernetes keep both in the same project, with the same login and roles.
Add-ons kept up to date
Update a cluster’s add-ons from the console instead of tracking each chart by hand.
Images close to the cluster
Browse container registries, repositories and tags next to the clusters that pull them.
Project roles
OIDC login, invitations and project roles decide who can change a cluster.
Questions.
Can I use kubectl and Helm?
Yes. Download the cluster’s kubeconfig and work with the tools you already use. [TBD: confirm kubeconfig access]
Which accelerators can a node pool use?
CPU machines today, with GPU and NPU machines to come. Tell us what you need and we’ll confirm what is available.
Can it run on our own infrastructure?
Yes. Supergate can run the platform, clusters included, on your infrastructure so your data stays there.
More solutions.
- Managed clustersManaged SlurmSlurm clusters for batch training and HPC jobs, the way researchers already work.
- Life sciencesBio pipelinesProtein design, sequence search and structure tools, run as pipelines without setting up each one.
- AI inferenceToken FactoryOpen models served through one API, priced per token.