Skip to main content

FPT Kubernetes Engine with GPU

FPT Cloud provides Kubernetes with NVIDIA GPU support, featuring the following key capabilities:

  • Flexible GPU configuration with multiple GPU types and optional GPU memory, applied per Worker Group.
  • Automatic GPU resource management and allocation in Kubernetes using NVIDIA Operator.
  • GPU visualization and monitoring with NVIDIA DCGM.
  • Automatic Container/Node scaling with Autoscaler when applications increase or decrease GPU resource usage.
  • GPU sharing support with Multi-Instance mechanism to optimize GPU resource utilization and cost.

FPT Cloud uses the NVIDIA GPU Operator, which provides tools to automatically manage all software components required to use GPUs on Kubernetes. The GPU Operator allows users to use GPU resources just like CPU resources in a Kubernetes cluster. Operator components include:

  • NVIDIA Drivers (CUDA, MIG, …)
  • NVIDIA Device Plugin
  • NVIDIA Container Toolkit
  • NVIDIA GPU Feature Discovery
  • NVIDIA Data Center GPU Manager (Monitoring)

In the Hanoi and Saigon regions, FPT Cloud currently supports Kubernetes with Nvidia A30 GPUs with the following MIG profiles:

No.GPU A30 ProfileStrategyNumber instanceInstance resource
1all-1g.6gbsingle41g.6gb
2all-2g.12gbsingle22g.12gb
3all-balancedmixed21g.6gb
412g.12gb
5none (no label)none00 (Entire)

In the Hanoi 2 and Japan regions, FPT Cloud currently supports Kubernetes with Nvidia H100 and Nvidia H200 GPUs:

No.GPU H100 SXM5StrategyNumber instanceInstance resource
1all-1g.10gbsingle71g.10gb
2all-1g.20gbsingle41g.20gb
3all-2g.20gbsingle32g.20gb
4all-3g.40gbsingle23g.40gb
5all-4g.40gbsingle14g.40gb
6all-7g.80gbsingle17g.80gb
7all-balancedmixed2 / 1 / 11g.10gb / 2g.20gb / 3g.40gb
8none (no label)none00 (Entire)
No.GPU H200 SXM5StrategyNumber instanceInstance resource
1all-1g.18gbsingle71g.18gb
2all-1g.35gbsingle41g.35gb
3all-2g.25gbsingle32g.25gb
4all-3g.71gbsingle23g.71gb
5all-4g.71gbsingle14g.71gb
6all-7g.141gbsingle17g.141gb
7all-balancedmixed2 / 1 / 11g.18gb / 2g.35gb / 3g.71gb
8none (no label)none00 (Entire)

Example: If you select strategy single: all-1g.6gb, the A30 GPU card on the worker is divided into 4 MIG devices, each with logical GPU resources equal to 1/4 of the physical GPU and 6 GB GPU RAM. If you select strategy single: all-1g.10gb, the H100 GPU card on the worker is divided into 7 MIG devices, each with logical GPU resources equal to 1/7 of the physical GPU and 10 GB GPU RAM.

Note: MIG config applies to all GPU cards attached to the worker. The MIG strategy across all worker groups in the same cluster must be of the same type (single/mixed/none).