Lydie.cc — Solar AI/LLM Server
This project is a high-performance local AI server built using recycled and older hardware components to run Large Language Models (LLMs) entirely via solar power. The goal was to create a cost-effective, decentralized AI system that costs $0.00 per token to run.
Spec list:
- Intel Core i7-3930K CPU
- Gigabyte GA-X79-UD5 Motherboard
- 64GB DDR3 RAM (8 x 8GB G.Skill Ripjaws X F3-1600C9D-16GXM)
- 3 x NVIDIA GeForce RTX 2080 Ti Blower Edition GPUs (33GB Total VRAM)
- Corsair RM1000e 1000W Power Supply (Refurbished)
- 240GB SATA SSD (OS Boot Drive)
- 3 x 1TB SATA SSDs (AI Storage Array)
- Advantech Industrial Computer 510 Case
Software Stack:
- OS: Ubuntu Server 26.04 LTS (Headless)
- Networking: Netplan, NetBird VPN (for remote access)
- AI/LLM Software: Unsloth Studio, Hermes Agent, Docker
Performance & Power:
- Runs Qwen3.6-35B-A3B-UD-Q4_K_M
- Initially achieved ~91.5 tps, which is impressive for a DDR3 platform and a 35B model!
- Power Consumption: Idle: ~100W / Load: Pushes 550W+ across the three GPUs during inference.
Cost & History:
- Budget: The build cost approximately $771.49 out-of-pocket for GPUs and an extra $56 for DDR3 RAM sticks.
- Value: The system has a replicated value of ~$1,340 and would have cost roughly $7,700 if built in the DDR3 era.
I asked the Hermes Agent running on this server to make a website about itself: https://lydie.cc/purpleai
