LLM decoding is dominated by memory bandwidth, and 3090Ti and 4090 happen to hav... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

junrushao1994 on Aug 10, 2023 | parent | context | favorite | on: Making AMD GPUs competitive for LLM inference

LLM decoding is dominated by memory bandwidth, and 3090Ti and 4090 happen to have the identical theoretical memory bandwidth

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact