Intelligence Report

OpenAI says its AI models found up to 14% PPA gains in AMD's flagship GPU designs

OpenAI says work with AMD produced about a year of chip design progress in under five months, and Anthropic reports similar speedups on AMD hardware.

INSIGHTEX Editorial Desk

2 min read

OpenAI says its AI models found up to 14% PPA gains in AMD's flagship GPU designsIntelligence Report

What happened

OpenAI says its models helped AMD improve power, performance and chip area, measured together as PPA, by as much as 14 percent on AMD's flagship GPU intellectual property, The Decoder reported. The joint effort took under five months. OpenAI described the result as roughly a year of progress, achieved on designs that AMD's own engineers had spent years refining.

According to OpenAI, AMD has already built part of the resulting changes into a chip design that has gone to manufacturing. OpenAI also used the same approach on Jalapeño, its first custom chip built for inference. In experiments on open RISC-V processor cores, OpenAI's GPT-6.2 Astra model produced the best results among its model generations.

Separately, OpenAI is developing GPT-Synopsys with Synopsys, a model that draws on Synopsys's EDA tools with the goal of widening access to AI-assisted chip design. Anthropic is doing related work. Its Mythos model completed prompt processing for a model running on AMD MI355X GPUs in three days, a job Anthropic says usually takes six to twelve months.

Context

Leading AI developers are paying closer attention to the hardware their models run on. OpenAI is building Jalapeño, its first custom inference chip. Anthropic now runs Claude on four chip families to improve cost and performance per dollar. Both companies are also pointing their models at chip engineering itself, including work on AMD hardware.

Astra profile

SourceThe Decoder

The briefing

The week in tech, in one email.

The stories worth your time, every week. Unsubscribe anytime.