← Back to archive

Sohu AI chip promises huge speed for transformer models

Just heard about Sohu, this new AI chip that's crazy fast - like over 500k tokens per sec with Llama 70B. Basically, one Sohu server can do what 160 H100s would do. It’s super specialized though - only for transformer models, but that's pretty much what all major AI products use now. Thoughts?

Sohu AI chip promises huge speed for transformer models

Originally posted on X

PS: I'm pretty active on 𝕏 if you'd like to follow along

Archive

2026

2025

2024

2018