OpenAI says its models found up to 14% PPA gains in AMD's flagship GPU designs

OpenAI says model-rewritten circuit code went into an AMD design sent for manufacturing; the GPU figures are OpenAI's own and have not been independently audited.
Why it matters
OpenAI says its models also cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit of Jalapeño, its first custom inference chip, gains it says were proven in silicon. AMD verified single-digit percentage gains on key blocks of its next flagship CPU using model-written designs. On two open-source RISC-V cores, GPT-6.2 Astra scored highest of five model generations across 9,881 scored runs. OpenAI is also working with Synopsys on GPT-Synopsys, a model that operates chip-design tools.
Key facts
- Up to 14% PPA gain on AMD flagship GPU IP
- About a year of progress in under five months, OpenAI says
- Jalapeño: about 13% vector-unit and 10% matrix-unit area cut
- AMD verified single-digit gains on next flagship CPU blocks
- 9,881 scored RISC-V runs; GPT-6.2 Astra scored highest