Nvidia's CUDA libraries can be generic and not optimized for LLM inference

(github.com) by venkat_2811 | Jan 18, 2026 | 1 comments on HN