Why I'm traveling to India to build a dual-RTX 3090 rig for local LLMs
Ubunta · x · 2026-08-22
A blogger shared a cross-border deployment plan and cost analysis for building a local GPU server to run Qwen-3.8:
Hardware Selection:
- GPU: 2× used RTX 3090 (24GB) are cheaper in India, with room for future upgrades.
- CPU/RAM: Ryzen 5600 + AM4 + 64GB DDR4 is cheaper to buy in Germany.
- PSU/Mobo: Sized for dual GPUs to enable easy future scaling.
Operational Costs: Electricity costs in India are significantly lower than in Germany, even after accounting for air conditioning.
Network & Remote Management:
- Uses encrypted WireGuard/Tailscale tunnels for remote access.
- Tested upload speeds from India are acceptable; network is not the main hurdle.
- Uses a 2013 Mac (Linux) as the remote management box.
- Plans to add PiKVM + UPS to debug the rig from 6,500 km away.
More from coding & agent
- How to build a mobile dispatch system to control laptop apps via Claude — holihai · 2026-08-22
- Forcing understanding over copy-pasting recommended for learning — Al_Grigor · 2026-08-22
- Understand Anything: Turn Codebases into Interactive Knowledge Graphs — JeremyCMorgan · 2026-08-22
- A 10,000-ft View of AI Agent Architecture — _jaydeepkarale · 2026-08-22
- Building Voice Agents is Easy. Making Them Sound Human is Hard. — grahamhart_ · 2026-08-22
- Opus 5's planning never ends: claims done, then finds new issues every review — broot66 · 2026-08-22