OpenAI's AI Models Improve Chip Design with Up to 14% Performance Gains
Summarized by AI from reporting by OpenAI Blog, published under our editorial policy.
OpenAI's latest models optimized circuits in its custom Jalapeño chip and AMD's flagship GPU IP, achieving up to 14% performance improvements and reducing chip area. The results show AI-assisted chip design is advancing rapidly.

Key takeaways
- OpenAI's GPT-6 Astra model achieved up to 14% performance improvements on AMD's flagship GPU IP.
- The models reduced standard-cell area by about 13% in Jalapeño's vector unit and 10% in the matrix unit.
- GPT-6.2 Astra outperformed GPT-6.1 Sol and GPT-5.6 Sol in optimizing open-source RISC-V cores.
- AMD included some of OpenAI's model-generated design changes in a chip sent for manufacturing.
- Starting from written requirements allowed AI models to achieve higher scores than when improving existing RISC-V core designs.
OpenAI's advanced AI models have made substantial strides in chip design, achieving significant performance gains. The company's engineers used these models to optimize circuits in their custom inference chip, Jalapeño, and improve designs for AMD's flagship GPU IP. These advancements highlight the potential of AI-assisted chip design to accelerate innovation in the semiconductor industry.
How OpenAI's Models Improved the Jalapeño Chip
OpenAI's models, particularly GPT-6 Astra, demonstrated the ability to make expert-level improvements in chip design. For the Jalapeño chip, the models identified ways to reduce the area of standard cells in the vector unit by about 13% and in the matrix unit by about 10%. These optimizations were verified during the chip's silicon bring-up. The vector unit, which manages and executes parallel computing tasks, saw savings from floating-point adders and multipliers, conversion functions, reduction logic, and the special-function unit. The models explored both alternative logic implementations and how to divide computations across pipeline stages, finding opportunities to share hardware and reduce intermediate storage. In the matrix unit, an arithmetic component, improvements accumulated across floating-point adders, multipliers, and low-precision dot-product circuits. One multiplier redesign used about 15% less cell area in synthesis while reducing latency from two clock cycles to one. A separate adder redesign reduced cell area by about 15% while retaining its two-cycle latency. Further improvements to dot-product logic and pipeline organization contributed additional savings across the block.
Results on AMD's GPU and CPU Designs
OpenAI partnered with AMD on power, performance, and area (PPA) improvements and saw up to 14% improvement relative to heavily optimized baselines on AMD's flagship GPU IP. The model identified further improvements by rewriting the SystemVerilog RTL code that describes the chip's circuits. This represents roughly a year of progress achieved in less than five months. AMD has already included some of these changes in a recent design it sent for manufacturing, and added others to its production design files for use in future versions. AMD also verified single-digit percentage gains in PPA across key blocks of its upcoming next-generation flagship CPU using designs written by OpenAI's model.
Benchmarking AI on Open-Source RISC-V Cores
To measure the progression in chip design capabilities from one model generation to the next, OpenAI tested each model on two open-source 32-bit RISC-V cores: PicoRV32 and Ibex. The models were asked to increase throughput (instructions per second). GPT-6.2 Astra achieved the highest average score on both cores, outperforming GPT-6.1 Sol and GPT-5.6 Sol. The models also showed a higher average score when designing a core from written requirements compared to improving an existing design. The full evaluation included 9,881 scored runs across both tasks, both cores, and all token budgets.
Why This Matters for Everyday Technology
The advancements in AI-assisted chip design could lead to more powerful and efficient processors in everyday devices. Faster, more efficient chips mean longer battery life for smartphones, more powerful laptops, and improved performance for data centers. This could also reduce the cost of producing these chips, making advanced technology more accessible to consumers. As AI continues to play a larger role in chip design, we can expect even more innovative and efficient processors in the future.
Frequently asked
- How did OpenAI's models improve chip design?
- OpenAI's models rewrote SystemVerilog RTL code to optimize circuits, reducing standard-cell area in Jalapeño by up to 13% and achieving up to 14% performance gains on AMD's flagship GPU IP.
- What is the Jalapeño chip?
- Jalapeño is OpenAI's first custom inference chip, designed to accelerate AI workloads. The models optimized its vector and matrix units during development.
- Did AMD use any of OpenAI's model-generated designs?
- Yes, AMD included some of the model-generated changes in a recent design sent for manufacturing and added others to production design files for future versions.
- How did OpenAI measure model performance on chip design tasks?
- OpenAI tested models on two open-source RISC-V cores, measuring throughput (instructions per second) after each model attempted to optimize or design the core from written requirements.