SIAgentNews.com
← All news

OpenAI uses AI to improve AMD GPU designs and its own Jalapeño chip

OpenAI's models achieved up to 14% gains in power, performance, and area for AMD's flagship GPUs, accelerating a year of design work into five months.

· First reported by The Next Web

14%gains in power, performance, and area

The short version

  • OpenAI's AI models improved AMD's flagship GPU designs by up to 14% in power, performance, and area.
  • The AI-driven process compressed a year's worth of design progress into under five months.
  • OpenAI used a similar approach for its own Jalapeño chip, reducing the size of two key parts by 13% and 10%.
  • OpenAI is partnering with Synopsys to train a model that runs existing industry design software.
  • OpenAI's GPT-6.2 Astra model scored highest in tests for both improving designs and building processors from scratch.

Why it matters

This demonstrates that AI can move beyond generating text to solving complex physical engineering problems like chip architecture. It suggests a future where hardware development cycles are significantly shortened by automated optimization.

Optimizing AMD hardware

OpenAI recently shared details on how its AI models are being used to refine high-end hardware. The company reported that its models improved power, performance, and area by up to 14% on AMD’s flagship GPU designs. These three metrics are the primary trade-offs that chip designers must balance when creating new hardware.

The project was particularly notable because it targeted AMD’s heavily optimized designs. Despite the complexity of these existing architectures, OpenAI's AI was able to find improvements that were already sent for manufacturing. The company also noted single-digit gains in key parts of AMD’s next flagship CPU.

According to a blog post reported by The Next Web, the AI achieved these results in under five months. This represents a significant acceleration of the design process, as the company estimated that the same amount of progress would typically take a year of human engineering effort.

How the AI rewrites circuits

The technical process involved the AI model rewriting the code that describes a chip’s circuits. Rather than just suggesting high-level ideas, the model interacted directly with the underlying specifications of the hardware. This allowed the AI to propose specific structural changes to the silicon layout.

To ensure these changes were functional, OpenAI used Synopsys software to check every modification. This verification step ensures that the AI-generated designs do not break the intended functionality of the chip. By combining generative capabilities with industry-standard verification tools, the company can validate its suggestions in real-time.

OpenAI also tested this methodology on its own Jalapeño chip, which is the company's first in-house processor. The AI-suggested changes successfully reduced the size of two key parts of the chip by approximately 13% and 10%. These improvements were eventually integrated into the finished silicon of the hardware.

Training specialized models

OpenAI is moving toward a more integrated approach by training a new model called GPT-Synopsys. This project is a collaboration with Synopsys, a company that is currently refocusing its business on AI-driven design. The goal is to create a model that can run the specific design tools that engineers already use in their daily workflows.

This move signals a shift toward making AI a standard part of the engineering toolkit rather than a standalone experiment. By training on the specific software used by professionals, the models can provide more practical assistance. This could eventually allow engineers to automate repetitive parts of the chip design cycle.

The company has already begun testing different versions of its models on these tasks. OpenAI tested five generations of its models on designing two small open-source processors. The newest model, GPT-6.2 Astra, performed the best in both improving existing designs and building new processors based on written requirements.

Human oversight and trust

While the results are impressive, the transition to AI-driven hardware design still faces hurdles regarding trust and reliability. OpenAI’s own chip chief, Richard Ho, recently expressed a cautious view on the current state of the technology. He stated that he still does not fully trust AI design changes, meaning every suggestion must be manually checked by humans.

This human-in-the-loop requirement is a standard safety measure in high-stakes engineering. Because a mistake in a chip design can cost millions of dollars and months of manufacturing time, engineers remain the final authority. The AI currently serves as a powerful assistant that identifies optimizations that humans might overlook.

The success of these tests is already attracting interest from the broader tech ecosystem. For example, the startup Architect Labs recently raised $24M in June. Their goal is to help various companies design custom chips using the same AI-driven methodologies that OpenAI is currently pioneering.

The future of silicon

The ability of AI to optimize physical hardware suggests a new era of rapid iteration for electronics. As models become more proficient at understanding circuit code, the time between a concept and a manufactured chip could shrink further. This could lead to faster updates for GPUs and CPUs across the industry.

OpenAI's work on the Jalapeño chip is a prime example of this trend. The chip moved from the initial design phase to manufacturing in just nine months. By using AI to cut down the size of key components, the company can achieve better efficiency without sacrificing the power of the hardware.

Looking ahead, the partnership with Synopsys will be a key indicator of how these tools scale. If GPT-Synopsys can successfully run standard industry tools, it may become the primary way engineers explore new chip architectures. The focus will likely shift from whether AI can design chips to how quickly it can do so while maintaining human-verified accuracy.

What to watch

  • Whether GPT-Synopsys becomes a standard tool for professional chip engineers.
  • How many more hardware companies adopt AI for optimizing their own custom silicon.
  • The final performance metrics of the Jalapeño chip once it is fully deployed.

Try it

Look up Synopsys's recent announcements regarding their AI-focused design tools to see how they are integrating with models like GPT-Synopsys.

Sources and links

First reported by The Next Web. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news