Skip to main content

Chapter Introduction

kstack: book: Centerpoint Home Lab chapter: AI & Automation page: Chapter Introduction tags: [ai, automation, ollama, n8n, cuda, gpu]

Overview

This chapter documents the AI inference, voice processing, and workflow automation services running on Centerpoint. All GPU-accelerated workloads leverage an NVIDIA GeForce RTX 5080 connected to the Mini PC via OcuLink, providing 16 GB of GDDR7 VRAM on the Blackwell architecture (Compute Capability 12.0).

GPU Hardware

Property Value
GPU NVIDIA GeForce RTX 5080
VRAM 16 GB GDDR7 (16,303 MiB)
Architecture Blackwell (Compute Capability 12.0)
Connection OcuLink (external GPU enclosure)
NVIDIA Driver 580.159.03
CUDA Version 12.9
Container Runtime nvidia (all GPU containers)

All containers that use the GPU are launched with runtime: nvidia and NVIDIA_VISIBLE_DEVICES=all. No device passthrough via --device /dev/dri is used — the Intel Arc / IPEX-LLM path has been retired.

Services in This Chapter

Service Container(s) GPU Purpose
Ollama ollama Yes Local LLM inference backend
Open Web UI open-webui No Chat interface for Ollama and OpenAI APIs
Faster-Whisper faster-whisper Yes Speech-to-text (Whisper large-v3-turbo)
Kokoro kokoro Yes Text-to-speech (TTS) API
Riffado riffado, riffado-db No AI audio podcast app (formerly OpenPlaud)
PaperlessAI paperless-ai No Autonomous document classification via Ollama
N8N n8n No Workflow automation platform
Open Notebook open-notebook, open-notebook-db No AI research notebook (SurrealDB backend)
MCP GitHub mcp-github-proxy, github-mcp-server No GitHub MCP server with OAuth 2.1

Compose Project

Most AI stack services (Ollama, Open Web UI, Kokoro, Faster-Whisper, Open Notebook, Riffado) are managed as a single Portainer compose project called ai-stack on a shared ai-internal bridge network plus traefik-net. N8N, PaperlessAI, and MCP GitHub are separate Portainer stacks.


Last Updated: 2026-06-16