🤜🤛 Synergy Plan: Starfleet & Risa
Objective: Balance workload based on hardware strengths to maximize performance and responsiveness.
📊 Hardware Comparison
| Feature | starfleet-compute (Mac Mini) | risa-recreation (HP EliteDesk) |
|---|---|---|
| CPU | i7-3720QM (Ivy Bridge, Gen 3) | i5-8500T (Coffee Lake, Gen 8) |
| Cores | 4 Cores / 8 Threads | 6 Cores / 6 Threads (True Cores) |
| GPU | HD 4000 (Legacy) | UHD 630 (Modern QuickSync) |
| Storage | ~1.5 TB Total (1TB HDD + 500GB SSD) | ~256 GB NVMe (Fast but Small) |
| RAM | 16 GB DDR3 | 16 GB DDR4 |
| Strength | Bulk Storage & Background Jobs | Media Processing & Responsiveness |
🏗️ The Roles
1. risa-recreation: "The Front of House" (User Experience)
This machine is newer, snappier, and has a modern GPU. It should run everything you interact with directly or watch.
- Role: Media Server & Gateway.
- Services:
- Plex / Jellyfin: (Moved from Starfleet). Relies on UHD 630 for flawless 4K/HEVC transcoding.
- Immich (Server + ML): Facial recognition will run 5x faster here than on the old Mac.
- Gaming: CS2 / Palworld Servers.
- The Gateway: AdGuard Home & Nginx Proxy Manager (ensures fast DNS/Web loading).
2. starfleet-compute: "The Engine Room" (Heavy Lifting)
This machine has more storage space and threads, making it ideal for large databases and "slow but steady" background crunching.
- Role: AI Core & Data Hoarder.
- Services:
- Ollama (AI): Keeps the "Local Brain" active.
- ChromaDB: Vector database for AI memory.
- Arr Stack (Radarr/Sonarr): These don't need a fast GPU, just storage access.
- Download Clients: qBittorrent / Sabnzbd.
- Local Backups: Use the extra 1TB drive as a staging area or backup target.
🚀 Execution Strategy
Phase 1: The Media Airlift (Risa)
- Goal: Move Plex and Immich to
risa-recreation. - Config: Mount
/mnt/vaultvia NFS. - Action: Deploy Docker stacks.
- Benefit: Immediate improvement in playback start times and seeking.
Phase 2: The AI Bunker (Starfleet)
- Goal: Dedicate resources to LLMs.
- Action: Ensure
Ollamahas dedicated access to the 16GB RAM (unburdened by Plex). - Optimization: Use the local SSD (
/mnt/ai_ssd) for ChromaDB and Model storage to speed up AI retrieval.
Phase 3: Network Glue
- Goal: Unified Access.
- Action: Run Nginx Proxy Manager on
risa. Pointplex.the-lal.nettorisa's IP (192.168.1.21). Pointai.the-lal.nettostarfleet(192.168.1.35).
✅ Pros & Cons
| Pros | Cons | |
|---|---|---|
| Speed | Massive UX Boost. Plex/Immich will feel "instant" on Gen 8 hardware. | Migration effort required. |
| AI | More RAM for Models. Freeing Starfleet from Plex prevents OOM crashes when running larger models (e.g. Llama-3-8b). | Network traffic increases (Plex reading files from NAS over network). |
| Efficiency | Right Tool for Job. QuickSync saves power during transcoding. |
🏁 Recommendation
Green Light. This split is logical. The Mac Mini is too old for modern video codecs (HEVC) but perfect for text-based AI and file management. The HP EliteDesk is a media powerhouse disguised as a tiny PC.