product · May 18, 2026
Hetzner Servers Power Self-Hosted LLM Setup with i7 CPU and 96GB RAM for 20-30 TPS
Share the canonical public link.
User @moishe_ee demonstrated a self-hosted LLM setup on a Hetzner i7 server with 96GB DDR4 RAM priced at 50 euros per month on May 17, 2026. The configuration runs ik_llama with qwen3.6-35B-A3B model achieving 20-30 tokens per second on 48K context length. The setup matches GPT-4 performance levels for local inference without expensive GPUs. @moishe_ee posted the video demonstration on May 17, 2026 with 128 likes and 4 reposts.
Spend governor blocked model creation: provider_circuit_open (lane=dev, provider=together)