Homelab serie -- The software 3/3

If you'd like to follow this series in detail, please visit the dedicated page below, where everything is neatly organized for easy reference:

Homelab – Tracking My Digital Journey

This computer sits beside the rack because I haven’t yet found a 4U rackmount case that allows me to properly consolidate and organize everything into a single enclosure. As highlighted in the system overview, the GPU is central to this setup’s performance needs.

System Overview

Operating System:

Distribution: Ubuntu 24.04.2 LTS (Noble Numbat)
Kernel: 6.8.0-55-generic

Hardware:

Motherboard: ASUSTeK TUF Z390-PLUS GAMING (Revision X.0x)
CPU: Intel Core i9-9900K (8-core, 64-bit, Coffee Lake architecture)
- Clock Speed: Averaging 800 MHz (boosts up to 5000 MHz)
- Cache: 512 KiB L1, 2 MiB L2, 16 MiB L3
Memory (RAM): 128 GiB (approximately 125.72 GiB available, 2.89 GiB used)
Graphics: Two NVIDIA GeForce RTX 3090 graphics cards
Storage:
- Main Drive: 1.82 TiB Samsung SSD 980 PRO 2TB (currently 9.5% used, 176.36 GiB)
- Partitioning:
  - / (Root): 1.79 TiB (9.6% used) - ext4 filesystem
  - /boot: 1.9 GiB (9.5% used) - ext4 filesystem
  - /boot/efi: 1.05 GiB (0.6% used) - vfat filesystem
Swap Space: 8 GiB file-based swap
Network:
- Wired connection (eno1) at 1000 Mbps (full duplex)

As detailed in several of my blog posts, I’ve built an AI setup using Ollama paired with OpenWebUI, which allows me to run large language models (LLMs) locally. For more technical details, see my write-up here :

Why This Machine?

I designed this system for continuous learning and research purposes. It functions as a "smarter personal assistant", helping me with coding experiments, troubleshooting technical issues, and exploring unconventional ideas. A key driver was my commitment to on-premises computing—keeping my data secure at home rather than relying on cloud services. Privacy is paramount in today’s digital landscape, so after years of outsourcing data to the cloud, I reversed course to retain full control over sensitive information.

Specific Use Cases

Health Monitoring
- My son’s diabetes management: The AI analyzes his glycemia data, identifying trends and patterns to avoid sifting through raw numbers or charts. This helps me focus on his well-being rather than getting lost in data noise.
Coding & Experimentation
- Accelerates prototyping by automating repetitive coding tasks, debugging suggestions, and even brainstorming novel solutions for my "weird ideas."
LLM Exploration
- Experimenting with cutting-edge models like:
  - Gemma 3-27B (Hugging Face )
  - Qwen 32B (Hugging Face ).
- The flexibility to download and test models locally keeps me ahead of the curve in AI development.

Technical Challenges

The system runs on dual NVIDIA RTX 3090 GPUs, which dominate PCIe slots—leaving no room for a high-speed 10GB SFP+ NIC. For now, I’m limited to a standard 1 Gb network card.

Vision & Inspiration

While not yet as sophisticated as Samantha in Her (a must-watch movie if you haven’t seen it!), this setup is a step toward integrating AI thoughtfully into daily life—prioritizing privacy, control, and innovation without compromising values.