platform-eng
NVIDIA DGX Multi Tenant Kubernetes Without Compromise
Deliver secure, isolated tenant clusters on NVIDIA DGX bare metal without VM overhead or weak namespace boundaries. vCluster virtualizes the K8s control plane itself.
Trusted by the fastest-growing AI cloud providers
Problem
DGX Multi-Tenancy Breaks at Scale
Standard Kubernetes forces impossible tradeoffs on expensive DGX hardware.
Solution
Tenant Clusters on DGX Bare Metal at Scale
vCluster virtualizes the Kubernetes control plane itself, running CNCF-certified tenant clusters as lightweight pods on NVIDIA DGX bare metal. Named in the NVIDIA DGX SuperPOD reference architecture, vCluster powers 100K+ GPU nodes across 50+ GPU clouds and Fortune 500 customers.
Built for NVIDIA DGX Multi Tenant Kubernetes
Every layer of the stack is purpose-built for isolated tenant clusters on high-density GPU infrastructure.
Tenant Control Planes on DGX Bare Metal
Each tenant gets their own CNCF-certified Kubernetes API server, etcd, and scheduler running as a lightweight pod on the DGX host cluster. No separate physical clusters required.
- Own API server per tenant
- Spun up in seconds
- Zero VM overhead on DGX
Fully Dedicated DGX GPU Nodes
Assign entire DGX physical nodes to a single tenant with their own CNI and CSI. No workloads from other tenants share the hardware, eliminating noisy-neighbor GPU contention.
- Dedicated physical DGX nodes
- Own CNI and CSI per tenant
- Complete hardware-level isolation
Kernel-Native Isolation Without VM Overhead
vNode (currently in private beta) provides container breakout protection using seccomp, cgroups, namespaces, and AppArmor per workload — preserving DGX GPU performance without hypervisor overhead.
- Prevents container breakout
- No hypervisor overhead
- No hypervisor overhead on GPU
Zero-Touch DGX Server Provisioning
PXE boot, OS installation, machine registration, and network automation for DGX servers handled automatically. Go from rack to production Kubernetes without manual intervention.
- PXE boot and OS install
- Full machine lifecycle management
- Zero-touch rack to production
Pre-Validated AI Stacks on DGX
Turn a DGX Kubernetes cluster into a production AI platform in minutes with pre-validated integrations for Run:AI, Ray, and Jupyter. Certified to work with tenant isolation out of the box.
- Run:AI, Ray, Jupyter ready
- Cluster to AI platform in minutes
- Certified with tenant isolation
Why vCluster
This isn’t a side project. Behind every vCluster deployment is 5+ years of deep K8s engineering, security hardening, and battle-tested infrastructure work at massive scale.
100K+
GPU Nodes Powered
50+
GPU Clouds & F500s
<45
Days to Launch
30K
GitHub Stars
Get Started in 3 Steps
1
Schedule a Demo
Talk to our team about your stack
2
Deploy vCluster
Deploy vCluster on your infra in minutes
3
Onboard Your Tenants
Go live with a hyperscaler-grade tenant experience in days
FAQs
How does vCluster provide tenant isolation on NVIDIA DGX systems?
vCluster virtualizes the Kubernetes control plane itself. Each tenant receives a fully isolated Kubernetes cluster with its own API server, etcd, scheduler, and RBAC — running as a lightweight pod on the DGX host cluster. This means tenants cannot see each other's nodes, pods, or platform internals. For stronger isolation, tenants can be assigned dedicated physical DGX nodes, and vNode (currently in private beta) adds kernel-native workload isolation without any hypervisor overhead.
Is vCluster compatible with NVIDIA DGX and the SuperPOD architecture?
Yes. vCluster is named in the NVIDIA DGX SuperPOD reference architecture. The platform is designed to run tenant clusters directly on bare metal DGX infrastructure, preserving GPU performance while delivering the isolated Kubernetes environments each tenant requires. vCluster powers over 100K GPU nodes in production.
Can we run NVIDIA DGX multi tenant Kubernetes without VM overhead?
Yes. vCluster tenant clusters run as lightweight processes on the host cluster, not inside traditional VMs. For workload-level isolation, vNode uses kernel-native primitives such as seccomp, cgroups, and AppArmor to prevent container breakouts without introducing a hypervisor layer. The result is strong tenant isolation at full bare metal DGX GPU performance.
How quickly can we launch a managed Kubernetes service on DGX hardware?
Boost Run launched a managed Kubernetes service in less than 45 days using vCluster, with zero new platform engineering hires. Lintasarta launched Indonesia's leading GPU cloud in 90 days with over 170 tenant clusters. vCluster's full stack covers bare metal provisioning through tenant cluster orchestration, eliminating months of custom platform engineering.
What isolation options are available for multi-tenant DGX deployments?
vCluster offers a flexible isolation spectrum. Shared nodes provide cost-efficient density with namespace and quota boundaries. Private nodes assign dedicated physical DGX hardware to a single tenant with their own CNI and CSI. Dedicated nodes eliminate noisy-neighbor GPU contention entirely. vNode (currently in private beta) adds kernel-native workload isolation at every tier, and tenant control planes can optionally run inside VMs for OS-level separation when compliance requires it.
Does each tenant get full Kubernetes cluster-admin rights on DGX?
Yes. Every tenant receives full cluster-admin within their own isolated Kubernetes environment. They can install CRDs, configure RBAC, and manage resources freely strongly isolated from other tenants and the underlying DGX host cluster, with blast radius contained to the individual tenant cluster. Each tenant cluster is CNCF-certified and 100% API compatible with standard Kubernetes.
Run Tenant Clusters on Your DGX Hardware
See how vCluster delivers isolated Kubernetes on NVIDIA DGX infrastructure.
