GPU Workloads

Run AI model inference, training, and GPU-accelerated workloads on dedicated NVIDIA hardware.

GPU Plan Specs

Spec	Value
GPU	RTX 4000 Ada (20GB)
CPU	2--8 vCPU
Memory	4--16 GB RAM
Use Cases	ML Inference, Training, CUDA

Getting Started

• Set your service plan to gpu when creating or updating
• Your container is scheduled on a GPU node with NVIDIA drivers and full 20 GB VRAM access
• CUDA libraries are available -- use GPU-enabled base images (e.g., nvidia/cuda:12.x)
• Your container runs with GPU acceleration and 20 GB VRAM

API

# Create a GPU service
curl -X POST https://railpush.com/api/v1/services \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-ml-model",
    "type": "web",
    "plan": "gpu",
    "image_url": "nvidia/cuda:12.6.3-runtime-ubuntu24.04",
    "start_command": "python serve.py",
    "port": 8000
  }'

# Or update an existing service to GPU
curl -X PATCH https://railpush.com/api/v1/services/<id> \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{"plan": "gpu"}'

MCP Tool

// Via MCP
{
  "tool": "update_service",
  "arguments": {
    "service_id": "<id>",
    "plan": "gpu"
  }
}