2 min read
Add as a preferred source on Google

OpenAI Reports AMD Chip Design Gains in Less Than Five Months

Some model-written changes are included in a design AMD sent for manufacturing; others are saved for future versions.

A verification probe touches the patterned surface of a chip prototype / TokenPost.ai
A verification probe touches the patterned surface of a chip prototype / TokenPost.ai

OpenAI’s models made roughly a year of progress on AMD GPU design work in less than five months, with gains of up to 14% across power, performance and area on the company’s flagship GPU intellectual property.

Some model-written changes are included in a design AMD sent for manufacturing; others are saved in production files for future versions. OpenAI did not name the AMD products or break down the gains across the three measures.

AMD confirmed single-digit percentage gains on several key blocks of its next-generation flagship CPU. OpenAI described the CPU results separately from the GPU figures.

The models rewrote SystemVerilog, a language used to describe chip circuitry. Synopsys verification tools checked each rewrite to confirm that the chip block’s behavior remained intact.

OpenAI directed GPT-o3 mini at Jalapeño, a custom inference chip developed with Broadcom, in early 2025. The company said proposed changes reduced standard-cell area by about 13% in the chip’s vector unit and 10% in its matrix unit.

OpenAI is also building a specialized chip-design model with Synopsys to operate electronic design automation tools directly. Its Preparedness Framework treats AI self-improvement as a risk area. OpenAI has committed to halt development if a model reaches the Critical self-improvement threshold, at least until appropriate safeguards are in place. The company warns that faster capability gains may create challenges for human control of AI systems.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…