Managed K8s with GPU Virtual Machine Overview
Managed K8s with GPU VM gives you a Kubernetes cluster with GPU-backed worker nodes on demand, without provisioning dedicated bare metal — FPT Cloud provisions GPU virtual machines and manages the control plane, with:
- Flexible GPU configuration — multiple GPU types and optional GPU memory sizing, applied per Worker Group.
- Automated management and provisioning of GPU resources with the NVIDIA GPU Operator.
- GPU visualization and monitoring with NVIDIA DCGM.
- Automatic node/container scaling with Autoscaler as GPU demand changes.
- GPU sharing with the Multi-Instance GPU (MIG) mechanism, to optimize GPU resource usage and cost.
In the Hanoi 2 and Japan regions, FPT Cloud currently supports Kubernetes using NVIDIA H100 and H200 GPUs.
If this is your first time using this service, complete the tasks in Initial setup before you create a cluster.
How GPU Operator works
FPT Cloud uses the NVIDIA GPU Operator to automatically manage the software components a Kubernetes cluster needs to use GPUs — so you can use GPU resources the same way you use CPU resources. The Operator installs and manages:
- NVIDIA Drivers (CUDA, MIG, and similar)
- NVIDIA Device Plugin
- NVIDIA Container Toolkit
- NVIDIA GPU Feature Discovery
- NVIDIA Data Center GPU Manager (monitoring)
MIG configuration reference
Multi-Instance GPU (MIG) splits a physical GPU into multiple smaller, isolated instances that Pods can share.
MIG configuration applies to every card attached to the worker. All worker groups in the same cluster must use the same MIG strategy type (single/mixed/none).
GPU H100 SXM5
| # | Label | Strategy | Instance count | Instance resource |
|---|---|---|---|---|
| 1 | all-1g.10gb | single | 7 | 1g.10gb |
| 2 | all-1g.20gb | single | 4 | 1g.20gb |
| 3 | all-2g.20gb | single | 3 | 2g.20gb |
| 4 | all-3g.40gb | single | 2 | 3g.40gb |
| 5 | all-4g.40gb | single | 1 | 4g.40gb |
| 6 | all-7g.80gb | single | 1 | 7g.80gb |
| 7 | all-balanced | mixed | 2 / 1 / 1 | 1g.10gb / 2g.20gb / 3g.40gb |
| 8 | none (no label) | none | 0 | 0 (entire GPU) |
GPU H200 SXM5
| # | Label | Strategy | Instance count | Instance resource |
|---|---|---|---|---|
| 1 | all-1g.18gb | single | 7 | 1g.18gb |
| 2 | all-1g.35gb | single | 4 | 1g.35gb |
| 3 | all-2g.25gb | single | 3 | 2g.25gb |
| 4 | all-3g.71gb | single | 2 | 3g.71gb |
| 5 | all-4g.71gb | single | 1 | 4g.71gb |
| 6 | all-7g.141gb | single | 1 | 7g.141gb |
| 7 | all-balanced | mixed | 2 / 1 / 1 | 1g.18gb / 2g.35gb / 3g.71gb |
| 8 | none (no label) | none | 0 | 0 (entire GPU) |
For example, selecting the single strategy with label all-1g.10gb splits the H100 GPU card on that worker into 7 MIG devices, each with 1/7 of the physical GPU's logical resources and 10GB of GPU RAM.
Terminology
| Term | Definition |
|---|---|
| K8s | Kubernetes |
| FKE | FPT Kubernetes Engine |
| D-FKE | Dedicated FPT Kubernetes Engine |
| M-FKE | Managed FPT Kubernetes Engine |
| Master Node | Node containing control plane components |
| Worker nodes | Nodes used for executing workloads |
| Automatic scaling of nodes | Automatic scaling of worker nodes (increase/decrease) |
| K8s cluster | A collection of nodes (VMs) configured as a Kubernetes cluster |
| NFS persistent storage | A persistent storage partition on NFS |
| Pod | The smallest unit managed by Kubernetes — contains one or more containers |
| Pod network | The network/subnet used to assign IP addresses to Pods |
| Service network | The network/subnet used to assign IP addresses to services |