What happened
OpenAI said in a blog post on Wednesday that its AI models produced gains as large as 14% across power, performance and chip area on AMD's flagship graphics chip designs, measured against designs AMD had already optimised extensively. Some of the changes are in a chip that has gone to manufacturing, TNW | Artificial-intelligence reported.
According to OpenAI, the model edited the code that defines a chip's circuitry and relied on Synopsys software to confirm each edit still functioned. The company described the result as roughly a year of progress achieved in less than five months. It also said AMD confirmed single-digit improvements in key sections of its upcoming flagship CPU.
OpenAI also tested five successive model generations on two compact open-source processor designs. GPT-6.2 Astra, the latest, came out on top at both refining existing designs and creating one from a written specification. OpenAI said the latter task was simpler, since models could drop features the originals had. It is now training a model called GPT-Synopsys with Synopsys to run the existing tools chip engineers rely on.
Context
OpenAI applied the method to Jalapeño, the first chip it built in-house, which reached production nine months after design began. Model-proposed edits cut two key components by roughly 13% and 10% and held up in the final silicon. Architect Labs raised $24M in June for AI-assisted custom chip design. OpenAI chip chief Richard Ho said last month that he does not yet fully trust AI-made design changes, so each is verified.
The briefing
The week in tech, in one email.
The stories worth your time, every week. Unsubscribe anytime.
