Skip to main content

GPU Cluster Overview

GPU Cluster gives you a managed Kubernetes cluster with GPU-backed worker nodes, so you can run AI/ML training, inference, and other GPU-accelerated workloads without operating the control plane yourself. FPT Cloud manages the control plane; you manage the worker nodes and what runs on them.

Two worker infrastructure options are available:

  • Managed K8s with Metal Cloud — worker nodes run on dedicated bare metal GPU servers (currently the flavor exposed as Managed GPU cluster on the Portal, under AI infrastructure).
  • Managed K8s with GPU VM — worker nodes run on GPU-backed virtual machines.

Both options integrate GPU Operator (driver and device management), Network Operator (GPUDirect RDMA), Slurm Operator, and vGPU Scheduler, and support cluster/worker-level autoscaling.

See the FAQ for common questions about regions, cluster configuration, compatibility, and scaling behavior.