A Guth Labs publication ↗
Back to News

OpenAI reports AI-assisted design improvements in AMD and Jalapeño chips

Blank brushed-metal chip on a green glass wafer with fine routing lines and one gold pathway
AI-generated editorial illustration

OpenAI published new chip-design results on November 18, saying its models helped improve AMD hardware designs and its own Jalapeño inference chip.

The company reports improvements of up to 14% across power, performance and area relative to heavily optimized baselines for AMD’s flagship GPU intellectual property. OpenAI says AMD has included some changes in a design sent for manufacturing and others in production design files for future versions. These are company-reported design results, not a claim that an entire shipping GPU runs 14% faster.

For Jalapeño, OpenAI estimates model-proposed changes reduced standard-cell area by about 13% in the vector unit and 10% in the matrix unit, with the optimizations validated during silicon bring-up.

The models rewrote circuit descriptions, while Synopsys tools checked whether the changes preserved behavior and measured engineering tradeoffs. OpenAI also tested successive models on two open-source RISC-V processor designs.

The results illustrate AI moving beyond writing software into hardware optimization, with verification tools and engineers still central to deciding which changes can enter production.

Sources and citations

Revision history

Published .