128GB local LLM server forges 80M tokens a month on just $6 of electricity
No-Fuel-9202 · reddit · 2026-09-26
A Reddit user details running a 128GB GMKtec X2 as an always-on local LLM server: on balanced performance settings it generates roughly 80 million Qwen3.8 Flash Next tokens per month for about $6 in electricity, while their regular usage is around 30 million tokens monthly. They also used the pi.dev coding agent's autoresearch extension to squeeze performance out of their astrometry app. With optimization work running dry, they're weighing ideas like expanding Karpathy's wiki or RAG-ing their codebase for the idle capacity — and asking the community for better uses.
More from coding & agent
- Sydney Codex Community Buildathon set for Nov 14 with Cerebras-backed ultrafast compute track — paw_lean · 2026-11-11
- shadcn's improve hits 10k stars: let top models plan, cheap models execute code fixes — shadcn · 2026-11-11
- OpenAI open-sources Codex Security CLI and SDK for AI-driven vulnerability scanning and fixing — openai · 2026-11-11
- Claude Opus 5.5 runs an architecture search and finds AttnRes-Canon, a new top Transformer design — SonglinYang4 · 2026-11-11
- 5 research habits worth knowing: read papers backwards, reproduce first, wire coding agents via MCP — cneuralnetwork · 2026-11-11
- 'Adam can now do CAE': AI agent showcase in engineering simulation draws buzz — burhop · 2026-11-11