OpenAI models helped AMD cut GPU design gains of up to 14% in PPA
The work compressed roughly a year of conventional chip design into under five months, and AMD has reportedly folded some changes into designs headed for manufacturing.
OpenAI reports that its frontier models, used on register-transfer-level hardware descriptions, delivered up to a 14% improvement in power, performance and area for an AMD flagship GPU, with single-digit gains on a next-generation CPU. The company says the work condensed roughly a year of conventional chip-design effort into under five months. AMD has reportedly folded some of these changes into designs now headed to manufacturing. In a separate benchmark on small open RISC-V cores, GPT-6.2 Astra scored highest on optimisation while keeping cost lowest, and the models did better when told to redesign a core from its specification than when asked to tune existing code.
Why it matters
The results suggest frontier AI models can take a real share of hardware design work, especially when asked to redesign blocks from specifications rather than tune existing code. For readers following the chip industry, it is a sign that the speed of chip development may change, though the evidence so far comes from OpenAI's own account and a small benchmark on open RISC-V cores, and AMD's use of the findings is only reported rather than confirmed.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.