Overview
GPU Container is a managed container platform on FPT AI Factory that gives you direct access to NVIDIA H100 GPUs. You select the number of GPUs, pick a ready-to-run template (or provide your own container image), expose the ports your application needs, and the platform handles provisioning.
Use cases
| Use case | Description |
|---|---|
| Jupyter Notebook | Remote development environment for AI/data scientists without local hardware limitations |
| Inference | High-throughput model inference using GPU resources with engines like vLLM and Ollama |
| Fine-tuning | Efficient fine-tuning of pre-trained models with GPU resources using PyTorch, CUDA, and NVIDIA Container Toolkit |
| Code Server | Cloud-based VS Code with GPU to train, test, and debug AI models remotely with full IDE capabilities |
Key concepts
GPU Instance — The compute unit for a GPU Container. Each instance is a fixed bundle of H100 GPUs, CPU cores, RAM, and temporary disk. You choose the size at creation time; it cannot be resized after the container is created.
Temporary disk — Fast local NVMe storage bundled with every GPU instance. Data on the temporary disk is lost when the container is stopped or deleted.
Persistent disk — Optional block storage you can attach at container creation time. Data persists across stop/start cycles. You set the capacity (1–3,000 GB) and the mount path inside the container.
Template — A pre-built container image provided by FPT. Templates include Jupyter Notebook, Code Server, Triton Inference Server, and vLLM variants. You can also use any image from a public or private registry.
Network ports — HTTP and TCP ports you expose from the container. The platform assigns a public URL for each HTTP port. TCP ports are accessible via SSH or direct TCP connections.
GPU instance sizes
All instances use NVIDIA H100 GPUs in the Ha Noi region.
| Instance | GPU | GPU VRAM | CPU | RAM | Temp disk |
|---|---|---|---|---|---|
| 1×H100 | 1 | 80 GB | 15 cores | 250 GB | 1 TB |
| 2×H100 | 2 | 160 GB | 30 cores | 500 GB | 2 TB |
| 3×H100 | 3 | 240 GB | 45 cores | 750 GB | 3 TB |
| 4×H100 | 4 | 320 GB | 60 cores | 1,000 GB | 4 TB |
| 5×H100 | 5 | 400 GB | 75 cores | 1,250 GB | 5 TB |
| 6×H100 | 6 | 480 GB | 90 cores | 1,500 GB | 6 TB |
| 7×H100 | 7 | 560 GB | 105 cores | 1,750 GB | 7 TB |
| 8×H100 | 8 | 640 GB | 120 cores | 2,000 GB | 8 TB |