OpenAI models improved AMD GPU design by up to 14 per cent

Share

OpenAI says its AI models rewrote parts of AMD’s flagship GPU architecture and found gains of up to 14 per cent in PPA, the main efficiency measure in chip design, over heavily optimized baseline designs. In a blog post, OpenAI said the collaboration delivered roughly a year of chip-design progress in less than five months, and AMD has put some of the model-generated changes into a design recently sent for manufacturing.

PPA stands for power, performance and area: how much energy a circuit draws, how fast it works and how much silicon it takes up. The models worked on SystemVerilog Register Transfer Level (RTL) code, which describes how a chip’s circuits carry out computations and move data. They explored alternative circuit implementations and pipeline arrangements, looking for ways to improve performance while reducing hardware requirements.

OpenAI said the models operated at the level of an expert microarchitect. They used Synopsys electronic design automation tools to check that each modification preserved the original circuit’s functionality and to measure its effect on PPA. Greg Brockman, OpenAI’s president, said in a post on X that the work had delivered “roughly an additional year of progress” on AMD’s chip in less than five months, and described the results as “unprecedented” at the frontier of chip design.

AMD independently verified the gains. It also confirmed single-digit percentage PPA improvements across key blocks of its upcoming next-generation flagship CPU, using designs generated by OpenAI’s models. Changes that did not make it into the taped-out GPU design have been added to AMD’s production design files for future versions. The figures published so far come from the two companies, which are commercial partners, and both expect further improvements in GPU and CPU designs as the models keep optimizing against existing baselines.

The chip design work builds on a multi-year agreement announced in October 2025, under which OpenAI committed to deploying up to six gigawatts of AMD Instinct GPUs. The first one-gigawatt deployment of MI450 GPUs is scheduled for the second half of 2026. AMD issued OpenAI a warrant for up to 160 million shares, with vesting tied to deployment and commercial milestones among others.

At AMD’s Advancing AI event in July, the companies outlined plans to optimize GPT-class workloads on AMD’s MI455X GPUs and Helios rack-scale systems using OpenAI’s Triton framework with AMD’s ROCm software. OpenAI expects to begin bringing Helios systems online in the fourth quarter of 2026, with deployments accelerating through 2027.

Source

OpenAI, AMD Use AI to Cut Chip Design Process From a Year to Under 5 Months

Scott Holmes
Scott Holmes
Scott Holmes is the Founder and Editor of InsightTrack AI, a Canadian publication covering artificial intelligence news, governance, security, and infrastructure. Based in Ontario, Canada, he brings more than 20 years of technology experience, including at Ericsson Canada, and holds PMP, CCNA, ITIL v3 Foundations, and Six Sigma certifications. His areas of expertise include AI governance, telecommunications, critical infrastructure, cybersecurity, and automation.

Table of contents [hide]

Read more

Local News