OpenAI said on November 18 that its AI models identified up to 14% improvement in power, performance, and area on AMD's flagship GPU intellectual property, measured against baselines the company described as heavily optimized by AMD engineers over years of work4. Some of the changes are in a design AMD has sent for manufacturing1.
The 14% figure is a best case, not an average across the design. OpenAI said the work took less than five months and represented roughly a year of normal progress.
OpenAI referenced prior work shared in August in which its engineers used frontier models for chip design. The November 18 post and accompanying research detail mark the company's most specific public claim to date on AI-driven hardware optimization results2. The post drew 203 points and 137 comments on Hacker News.