Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU from Hacker News on 2026-02-21 20:57 (#73R17) Comments
