OpenAI reports production chip gains from model-written RTL
OpenAI says model-proposed changes reduced standard-cell area by about 13% in Jalapeño’s vector unit and 10% in its matrix unit, while AMD reports up to 14% improvement over optimized GPU IP baselines and single-digit gains on next-generation CPU blocks.
Why it matters: The report moves agentic chip design from toy benchmarks toward silicon-proven and manufacturing-bound changes, while keeping expert verification and EDA equivalence checks in the loop.