Running a Local LLM on RTX 3090 Costs €0.06‑€0.30 per Million Tokens
I measured GPU electricity for eight open‑source LLMs on a single RTX 3090. The per‑million‑token cost ranged from about €0.06 for the mid‑sized Falcon‑7B to €0.30 for the heavyweight LLaMA‑13B, overturning the assumption that smaller models are always cheaper. These numbers let you budget inference runs with real‑world power costs.