SemiAnalysis: NVIDIA Leads AMD by Up to 5x in Inference Cost Efficiency

SemiAnalysis reports that NVIDIA's inference cost efficiency reaches up to 5x that of AMD at 150 tok/s/user, attributing the gap largely to AMD's weaker software stack. The firm also released AgentX 1.0 alongside the findings.

2026-08-25 ~ 2026-08-25 · 2 related posts