Get your GPUs working sooner
GPU capacity is only the beginning. Before your first workload runs, you need a control plane, connected worker nodes, GPU drivers, networking, and cluster access.
Hyperbolic brings those steps into one workflow. Choose your GPU capacity in the console, launch your cluster, and connect with kubectl when it’s ready.
Leave control-plane operations to us
Hyperbolic runs a highly available control plane for each cluster, with automated backups, and managed upgrades are coming soon. Your team can focus on deploying workloads without operating control-plane nodes or etcd.
Start with GPU-ready workers
The NVIDIA GPU Operator and Network Operator are installed during provisioning. GPU drivers and RDMA interfaces are available to your pods, so you can get straight to configuring your workloads.
Keep your Kubernetes workflow
Use kubectl, Kubernetes APIs, and your existing deployment tooling. Manage your own applications, namespaces, permissions, and network policies.
GPU infrastructure and Kubernetes, in one place
01
Configure the capacity your workload needs
Choose your GPU type, node count, and region from the available capacity in the console. Worker nodes within a cluster share the same region and interconnect.
02
Adjust capacity through node groups
See each node group’s GPU type, size, and status on the Compute tab. Add or remove nodes there as your compute requirements change.
03
Run across multiple GPU nodes
Use your framework’s NCCL launcher for multi-node training. The preinstalled NVIDIA Network Operator exposes RDMA interfaces to pods, and one-click Kueue adds job queueing and gang scheduling, so every part of a multi-node job starts together.
04
Connect with your own identity
Download a per-cluster kubeconfig and sign in through Hyperbolic. Access uses short-lived credentials tied to individual users. Organization admins receive cluster-admin access, while other members receive edit access.
05
See what’s happening on your cluster
Follow provisioning progress, inspect pods and node events, and view live GPU telemetry from the console. Built-in monitoring with Prometheus and Grafana includes GPU dashboards.
06
Keep control of your environment
Configure RBAC, namespaces, network policies, and add-ons to fit your team. Hyperbolic prepares the infrastructure; you decide what runs on it and how it operates.
From GPU selection to your first workload
Choose your GPU capacity
Open Kubernetes in the Hyperbolic console. Select your GPU type, region, and node count, then review your configuration before launching.
Let Hyperbolic prepare the cluster
Hyperbolic provisions the workers, joins them to the hosted control plane, and installs the GPU and networking add-ons. Follow each stage in the console until your cluster is ready.
Connect with kubectl
Install the authentication plugin once, download your cluster’s kubeconfig, and sign in through Hyperbolic. Check that your worker nodes are ready.
Deploy your workload
Apply your Kubernetes manifests, request GPUs for your containers, and inspect your pods and GPU activity from the console.
Connect with kubectl
# Once: install the sign-in plugin
kubectl krew install oidc-login
# Use your cluster's kubeconfig
export KUBECONFIG=~/Downloads/my-cluster.yaml
kubectl get nodes
Request GPUs in a pod
resources:
limits:
nvidia.com/gpu: 8
Your Managed Kubernetes questions, answered
Hyperbolic Managed Kubernetes is a service for running Kubernetes workloads on GPU instances with a hosted control plane operated by Hyperbolic. It combines cluster provisioning, GPU software setup, standard Kubernetes access, and platform visibility in one console.
Renting GPU instances gives you compute that you configure and operate. Managed Kubernetes adds a hosted Kubernetes control plane, joins your instances as worker nodes, and installs the initial GPU and networking components.
You can also rent interconnected GPU instances on Hyperbolic without Kubernetes.
Available GPU types, regions, and cluster sizes depend on current capacity. Check the console for configurations you can launch, or contact our team for larger deployments and specific capacity requirements.
Yes. The Compute tab lets you view your node groups and add or remove nodes. Capacity changes are managed through the console; automatic scaling is not part of the offering described here.
Yes. Connect through kubeconfig and use kubectl, Kubernetes APIs, and compatible deployment tools. Your team controls its workloads and in-cluster configuration.
Your kubeconfig uses Hyperbolic single sign-on to authenticate you with short-lived credentials. Organization admins receive cluster-admin access; other members receive edit access. You can extend Kubernetes RBAC for your team and service accounts.
Pods can use fast node-local NVMe through local persistent volumes. This storage doesn’t survive node replacement, so keep checkpoints in persistent storage.
Shared network storage is available on request. Contact our team.
Yes. Worker nodes have public IPs, and services can be exposed through the supported Kubernetes networking options after the required inbound ports are opened. Your team manages the serving application and its network configuration.
The console provides live GPU telemetry, node-level pods and events, and cluster status. Built-in monitoring with Prometheus and Grafana includes GPU dashboards, and you can add your own logging and alerting tools inside the cluster.
Termination removes the control plane, releases all worker nodes, and stops billing. Data remaining on the cluster is not recoverable, so save anything you need to persistent storage first.
A cluster is also terminated if its account balance reaches zero. Auto Top-Up helps prevent this.
