SemiAnalysis: NVIDIA holds up to 5x cost efficiency lead over AMD

rohanpaul_ai · x · 2026-08-25

A new SemiAnalysis report reveals that NVIDIA achieves up to 5x better cost efficiency than AMD at 150 tokens per second per user. The report also introduces AgentX 1.0, the first open-source multi-turn agentic coding benchmark with 1M context length. Testing across chips like GB300 and MI355 confirms NVIDIA's sustained software and performance moat in agentic workloads.

Related event: SemiAnalysis: NVIDIA Leads AMD by Up to 5x in Inference Cost Efficiency(2 posts)→

Original post →

More from Infra

Infra channel →