Super HN

   Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU (github.com) 8 points by xaskasdf 2 hours ago