AI Models & Platforms
OpenAI Reports Model-Driven Gains in AMD Flagship Chip Designs

Key PointsOpenAI's models rewrote AMD chip RTL for up to 14% PPA gains on flagship GPU designs.
View more
- OpenAI's models rewrote AMD chip RTL for up to 14% PPA gains on flagship GPU designs.
- Model-proposed changes cut standard-cell area by about 13% in Jalapeño's warp engine.
- Specification-based scores may overstate capability because models omitted features like performance counters.
OpenAI reported on November 18, 2026 that its models achieved up to a 14% improvement in power, performance and area (PPA) relative to heavily optimized baselines on AMD flagship GPU intellectual property, with some circuit rewrites included in a design sent for manufacturing.
The gains came from rewriting SystemVerilog RTL, the code that describes a chip’s circuits. The model used Synopsys tools to verify that each rewrite preserved a block’s behavior and to assess PPA trade-offs. AMD added other changes to production design files for future versions. OpenAI said the GPU work represented roughly a year of progress achieved in less than five months, and that AMD verified single-digit percentage PPA gains across key blocks of an upcoming next-generation flagship CPU using model-written designs.
Circuit-Level Gains in Jalapeño
OpenAI started using GPT-4o to help design Jalapeño, its first custom inference chip, in early 2025. The team estimates that model-proposed changes reduced standard-cell area by about 13% in the warp engine, which manages and executes parallel computing tasks, and by 10% in the matrix unit, an arithmetic component. The models explored alternative logic implementations, ways to divide computations across pipeline stages, opportunities to share hardware and reductions in intermediate storage.
OpenAI’s detailed examples include a matrix-unit multiplier redesign that used about 15% less cell area in synthesis while reducing latency from two clock cycles to one. A separate adder redesign cut cell area by about 15% while retaining two-cycle latency. Further improvements to dot-product logic and pipeline organization contributed additional savings across the block, and the optimizations were proven in silicon during Jalapeño’s bring-up.
OpenAI’s August 25, 2026 silicon results reported that AI assistance helped Jalapeño move from initial design to tapeout in nine months.
RISC-V Tests Compare Five Model Generations
To measure how chip-design capability changed across generations, OpenAI tested GPT-6.2 Astra, GPT-6 Astra, GPT-6.1 Sol, GPT-5.6 Sol and GPT-5.5 on two open-source 32-bit RISC-V cores. PicoRV32 is a compact core designed for reprogrammable chips called FPGAs; OpenTitan uses Ibex as the embedded processor in its security chip. The evaluation covered both optimization of an existing core’s RTL and creation of a core from written requirements and interface templates, with 9,881 scored runs across the tasks, cores and token budgets.
For RTL optimization, each model received the original hardware description and was asked to increase throughput, the number of instructions executed per second. A design scored zero if it failed a simulation check or used more than 16 times the original core’s area. For passing designs, OpenAI took the geometric mean of instructions per clock across workloads hidden from the models and multiplied it by the maximum clock frequency estimated with Yosys. The best observed throughput for each core under the same constraints was assigned a score of 1.0, and the averages included zero-scoring runs.
At a comparable budget of about 112,000 tokens per run, OpenAI’s results gave GPT-6.2 Astra the highest average score on both cores in RTL optimization; GPT-6.1 Sol also outscored GPT-5.6 Sol on both. In specification-to-RTL design near the same budget, GPT-6.2 Astra recorded a mean normalized score above 0.7 on PicoRV32 and averaged about 0.75 on Ibex. All five models scored higher when designing from written requirements than when optimizing existing RTL on both cores.
Evaluation Limits and the GPT-Synopsys Plan
The stronger specification-based scores come with a qualification. OpenAI said models starting from written requirements could omit features found in the original cores, such as performance counters and additional instructions, that did little to improve throughput on the test workloads. Those omissions could explain part of the score advantage. Requiring the features, setting a tighter circuit-area limit and checking timing after physical synthesis would make the test more representative of production chip design.
To connect model development with established verification systems, OpenAI is partnering with Synopsys to train GPT-Synopsys to use electronic design automation (EDA) tools the way expert chip designers do. The work builds on a multi-year agreement announced on September 30, 2026, covering joint development, go-to-market collaboration and revenue sharing. Synopsys said engineers would delegate objectives from PPA optimization to timing and verification closure, with agents running tools, interpreting results, implementing changes and iterating toward verified outcomes for engineer review.
Synopsys said customer design data would not be used to train GPT-Synopsys and would be encrypted at rest and in transit, with configurable retention, audit and permission controls. Those protections and the requirement for engineer review address the realities of working on proprietary production designs. The research’s central challenge is joining model-driven design gains to trusted verification; GPT-Synopsys is intended to provide that connection across circuit optimization and broader EDA workflows.












