Key Notes
- OpenAI reports up to 14% improvement in power, performance and area on AMD GPU designs, rather than a whole-chip speed benchmark.
- AMD has included some proposed circuit changes in a design sent for manufacturing, according to OpenAI.
- The work uses Synopsys verification tools and builds on a previously announced partnership to develop a specialized chip-design model.
OpenAI says its models have improved parts of AMD‘s flagship GPU design by up to 14%, with some AI-proposed circuit changes already included in a design sent for manufacturing. The November 18 report offers a concrete example of AI contributing to the hardware on which future models will run.
The company reported gains against highly optimized starting designs in power, performance and area, the engineering measures known as PPA. It did not announce a 14% improvement in the speed of an entire commercial GPU. The result concerns circuit-design work, and OpenAI has not identified the product involved.
Editing the Circuits Behind the Chip
The models rewrote SystemVerilog RTL, the code describing a chip’s circuits. Synopsys software checked that proposed changes preserved the required behavior and evaluated their engineering trade-offs. OpenAI also reports single-digit PPA improvements in key blocks of AMD’s next flagship CPU.
For AMD, the distinction between a plausible suggestion and a usable design is central. A circuit change has to work with the rest of a processor and survive verification. Making one component smaller, for example, can be valuable without increasing the headline speed of the finished product.
The work extends the AI-assisted engineering behind OpenAI’s own Jalapeño inference chip, whose first performance results we covered in August. It is a different task from asking a coding assistant to write software that runs on a finished processor: here, the assistant is helping change the processor itself.
The Software Partnership Behind the Work
Synopsys and OpenAI announced a multiyear agreement on September 30 to develop GPT-Synopsys, a specialized model for semiconductor design. The partnership combines OpenAI’s models with Synopsys’ electronic design automation tools and includes joint commercial work and a revenue-sharing arrangement.
The intended workflow keeps engineers in the review process. They set design goals, while agents operate tools, interpret results and revise designs toward verified outcomes. Synopsys says the planned offering will combine compute, model access and software licenses on OpenAI-hosted infrastructure.
The companies also describe access controls, encryption and configurable retention for customer designs, with customer data excluded from model training. These are consequential requirements for a tool that would work on proprietary circuit descriptions rather than publicly available code.
What the Results Establish
OpenAI’s separate tests on two small RISC-V processor cores show why production claims and research benchmarks need different treatment. Models sometimes scored better when designing from written requirements, but those requirements allowed features from the original cores to be omitted. OpenAI acknowledges that tighter constraints would better reflect production engineering.
The manufacturing-stage AMD work is the more immediate development. Some proposed changes have progressed beyond an experiment, according to OpenAI. Their effect on a named product’s performance, cost and power consumption will need evidence from the finished hardware.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.