Etched Sohu vs. Nvidia: Transformer ASIC vs. GPU (2026) – Spheron Blog
Etched Sohu claims 500k tokens/sec on Llama 70B from an 8-chip server. What a transformer-only ASIC gives up versus H100, B200, and the Groq 3 LPU.
Summary
Etched Sohu’s new transformer ASIC promises impressive token processing speeds, but how does it stack up against Nvidia’s GPUs and other competitors? This comparison highlights the trade-offs in performance and efficiency between specialized chips and general-purpose GPUs.
Comments (4 of ) ⌛