OpenAI reports up to 14% design gains for AMD flagship GPUs
Some AI-suggested changes are in a chip sent for manufacturing, OpenAI said, while every proposed design change remains subject to checks.

OpenAI said its models delivered improvements of up to 14% across power, performance and area measures in AMD’s flagship GPU designs. Some of the modifications have been incorporated into a chip sent for manufacturing, the company said.
The models revised code defining the chips’ circuitry, with software from Synopsys checking whether the changes preserved functionality. OpenAI described the results as roughly a year of progress achieved in less than five months, despite AMD’s designs already being heavily optimized. According to OpenAI, AMD also confirmed single-digit improvements in important sections of its next flagship CPU.
For Jalapeño, OpenAI’s first internally developed chip, model-generated revisions reduced the size of two key sections by about 13% and 10%. OpenAI said those changes worked in the manufactured silicon. The chip took nine months to move from design into manufacturing.
Separate tests compared five model generations on two small open-source processors. GPT-6.2 Astra, the newest model, performed best at both modifying existing designs and creating processors from written specifications. OpenAI cautioned that the latter task was easier because models could omit features present in the original processors.
OpenAI and Synopsys are now training GPT-Synopsys to operate engineers’ existing design tools. OpenAI chip chief Richard Ho said he did not yet fully trust AI-generated design modifications, and each one undergoes checking.




