- Models rewrote SystemVerilog RTL to achieve up to 14% PPA improvement on AMD's flagship GPU IP
- Jalapeño's vector and matrix compute units cut standard-cell area by about 13% and 10%, validated on real silicon
- OpenAI is partnering with Synopsys to train GPT-Synopsys to operate EDA tools like a chip designer
AI CURATED HOT STORY · 11/18/2026
OpenAI Models Boost AMD Chip Design with 14% PPA Gain in Five Months
OpenAI has disclosed that its frontier models have deeply participated in chip design, achieving up to 14% PPA (power, performance, area) improvement on AMD's flagship GPU IP and compressing about a year of optimization progress into less than five months. Meanwhile, the models delivered expert-level optimizations on OpenAI's self-designed inference chip Jalapeño, reducing standard cell area of vector and matrix compute units by approximately 13% and 10%, respectively, with some improvements validated on actual silicon.
Technical details reveal that the models optimize circuit structures by rewriting SystemVerilog RTL, achieving non-trivial improvements in key modules such as floating-point adders and multipliers in Jalapeño, and achieving best results in RISC-V open-source core tests. OpenAI is extending the same methodology to AMD's third-party hardware, and some optimizations have entered tape-out designs, indicating that AI-assisted design has moved from experimentation to production. Additionally, OpenAI is collaborating with Synopsys to train a dedicated GPT-Synopsys model that enables the model to directly operate EDA tools for design and verification, driving end-to-end automation.
This progress marks a leap for AI in chip design from an auxiliary tool to a core productivity driver. For AMD, the additional optimization progress from AI could accelerate product iteration; for OpenAI, rapid optimization of its custom chip helps reduce inference costs. The collaboration with Synopsys may reshape the interaction paradigm of EDA toolchains, bringing an efficiency revolution to the chip design industry.
Hourly heat curve
36 fully observed hoursThe line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.
Coverage timeline
Total 4 reports · Latest first- IT之家 · IT业界T2·88 pts
- The DecoderT2·78 pts
- OpenAI models delivered up to 14% PPA gains on AMD's flagship GPU IP in under five months, equal to about a year of engineering progress
- Some optimizations were folded into a design sent to manufacturing, showing AI-assisted design reaching production
- The method was reused on OpenAI's custom inference chip Jalapeño, achieving best results on RISC-V core tests
- Hacker News AIT2·78 pts
- Models including GPT-4o delivered nontrivial PPA gains in Jalapeño's vector unit and other blocks
- OpenAI is extending the chip-design approach to AMD hardware for significant gains
- The Synopsys partnership yields GPT-Synopsys, embedding frontier models into mainstream EDA tooling
- OpenAI 官方博客T1·78 pts
- OpenAI frontier models achieved nontrivial PPA gains in Jalapeño's vector unit, validating AI participation in expert-level chip design
- The same approach is being extended to third-party AMD hardware, so gains are not limited to in-house chips
- A GPT-Synopsys model embeds frontier intelligence into existing EDA toolchains