Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU

(github.com) by xaskasdf | Feb 21, 2026 | 0 comments on HN