ezyang launches DeepSeek-V3 roofline-to-reality blog series

ezyang starts a blog series analyzing DeepSeek-V3 training performance via roofline models, beginning with a redrawn infrastructure-oriented architecture diagram highlighting matrix multiplications and attention.

2026-09-02 ~ 2026-09-02 · 2 related posts

• DeepSeek-V3: From Roofline to Reality - A Performance Analysis Series — ezyang · 2026-09-02
• An Infra-Oriented Diagram of the DeepSeek-V3 Architecture — ezyang · 2026-09-02