Hermes Vector

AI & SI

OpenAI Models Improve Chip Design in Jalapeño, AMD, and RISC-V Tests

OpenAI reports its models reduced cell area in its Jalapeño chip, helped AMD improve GPU and CPU designs, and scored highest on RISC-V core optimization tasks.

By Hermes Vector

Also in: Español · Português · Français · Deutsch · 中文

Checked against 1 sourceOpenAI
Illustration for: OpenAI Models Improve Chip Design in Jalapeño, AMD, and RISC-V Tests

In brief

  • OpenAI says model-proposed changes reduced standard-cell area by about 13% in Jalapeño’s vector unit and 10% in its matrix unit.
  • OpenAI reports up to 14% power, performance, and area improvement on AMD’s flagship GPU IP relative to heavily optimized baselines.
  • AMD verified single-digit percentage gains in power, performance, and area across key blocks of its upcoming next-generation flagship CPU using OpenAI model designs.
  • GPT-6.2 Astra had the highest average score on both PicoRV32 and Ibex RISC-V cores with about 112,000 tokens per run.

OpenAI Models Improve Chip Design in Jalapeño, AMD, and RISC-V Tests

OpenAI says its models are improving the chips that power AI, including its first custom inference chip, Jalapeño, and third-party hardware with AMD.

Jalapeño design improvements

OpenAI engineers started using GPT-4o in early 2025. The company says that early model found ways to make nontrivial power, performance, and area improvements. In the vector unit, the savings came from floating-point adders and multipliers, conversion functions, reduction logic, and the special-function unit. In the matrix unit, 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. Across the broader Jalapeño design effort, the team estimates that model-proposed changes reduced standard-cell area by about 13% in the vector unit and 10% in the matrix unit. These optimizations were silicon proven during Jalapeño’s silicon bring-up.

Collaboration with AMD

OpenAI has partnered with AMD on power, performance, and area improvements and seen up to 14% improvement relative to heavily optimized baselines on the flagship GPU IP. The model identified further improvements by rewriting the code that describes the chip’s circuits, SystemVerilog RTL. OpenAI says this represents roughly a year of progress achieved in less than 5 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 power, performance, and area across key blocks of its upcoming next-generation flagship CPU using designs written by OpenAI’s model.

RISC-V core optimization

To measure progression in chip design capabilities, OpenAI tested each model on two open-source 32-bit RISC-V cores: PicoRV32, a compact core designed for FPGAs, and OpenTitan, which uses Ibex as the embedded processor in its security chip. Each model was given the original hardware description for each core and asked to increase throughput. Designs scored zero if they failed a check or used more than 16 times the original core’s area. With a comparable budget of about 112,000 tokens per run, GPT-6.2 Astra had the highest average score on both cores. On the existing PicoRV32 core, GPT-6.2 Astra has the highest mean score at the larger comparable budgets. Around 112,000 tokens, its score is about twice GPT-6 Astra’s. On the existing Ibex core, GPT-6.2 Astra’s mean score continues to climb across the higher budgets it reaches.

For a second task, models were given written requirements and templates and asked to design a core from a specification. At about 112,000 tokens per run, all five models scored higher when they designed a core from written requirements than when they tried to improve the existing design. The full evaluation included 9,881 scored runs across both tasks, both cores, and all token budgets.

Why it matters

OpenAI’s results show that its models can propose changes to production chip designs and explore new processor architectures. The company is partnering with Synopsys to train GPT-Synopsys, a new model, to use Synopsys electronic design automation tools the way expert chip designers do.


Sources

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.

Questions readers ask

What chip design improvements did OpenAI achieve with its models?
OpenAI engineers used GPT-4o in early 2025 to reduce standard-cell area by about 13% in the vector unit and 10% in the matrix unit of the Jalapeño chip. These optimizations were silicon proven during Jalapeño’s silicon bring-up.
How has OpenAI collaborated with AMD on chip design?
OpenAI partnered with AMD to achieve up to 14% improvement on flagship GPU IP by rewriting SystemVerilog RTL code. AMD has included some of these changes in a recent design sent for manufacturing and verified single-digit percentage gains in its upcoming next-generation flagship CPU.
Which model performed best in RISC-V core optimization tests?
GPT-6.2 Astra had the highest average score on both PicoRV32 and OpenTitan cores, with a score about twice that of GPT-6 Astra at around 112,000 tokens. The full evaluation included 9,881 scored runs across both tasks, both cores, and all token budgets.
Why do these results matter for chip design?
The results show that OpenAI's models can propose changes to production chip designs and explore new processor architectures. The company is partnering with Synopsys to train GPT-Synopsys to use Synopsys electronic design automation tools like expert chip designers.

The day's checked news, by email

One short email a day with the top stories. Free, and you can unsubscribe in one click.

Or follow for the latest news:

How do you feel about this story?

Comments 0

No comments yet. Start the discussion.