# OpenAI helps AMD boost GPU efficiency by up to 14% in five months

> OpenAI’s AI models improved power, performance and area on AMD’s flagship GPU by about 14%, which AMD says equals a year's worth of engineering work.

Oossa · 2026-11-18 · https://oossa.com/en/openai-helps-amd-boost-gpu-efficiency-by-up-to-14-in-five-months

OpenAI announced that its AI models have been used to tune AMD’s flagship GPU design. In less than five months the models added up to a 14 percent improvement in power, performance and chip area – the three metrics AMD groups under the acronym PPA. AMD says that improvement matches roughly a year of work that its engineers would normally need to achieve on top of the design that has already been refined for years. Some of the changes have already been folded into a version of the chip that is now headed to the fab.

The work was done in partnership with AMD’s hardware team. OpenAI fed its GPT‑style models with layout and timing data, letting the AI suggest small tweaks that would reduce power draw, speed up computation, or shrink the silicon footprint. The company says the same approach helped on its own custom inference chip, called Jalapeño, where the GPT‑6.2 Astra model beat earlier versions on open‑source RISC‑V cores. OpenAI is also teaming with EDA vendor Synopsys on a project called GPT‑Synopsys, which aims to make AI‑assisted chip design tools available to a broader set of engineers.

## How other AI firms are using similar tricks

Anthropic, another AI lab, reports a similar speed‑up on its Mythos model. The model processed prompts for a design running on AMD’s MI355X GPUs in three days, a timeline Anthropic says is normally six to twelve months. Anthropic now runs its Claude assistant across four AMD chip families to balance cost and performance per dollar. These examples show that AI‑driven design is moving from research labs into real silicon production.

## Why this matters for the wider market

If AI can shave a year off the development cycle for high‑end GPUs, chip makers could bring more efficient hardware to customers faster and at lower cost. For gamers, data‑center operators, or anyone buying a graphics card, the benefit could be longer battery life on laptops, cheaper cloud compute, or smaller devices with the same performance. The exact savings will depend on how much of the AI‑generated tweaks make it into mass‑produced chips, a detail AMD has not disclosed yet.

## The facts

- OpenAI’s models improved AMD’s flagship GPU PPA by up to 14 percent in under five months.
- AMD says the gain equals about a year's worth of engineering progress on top of existing designs.
- Some of the AI‑suggested changes have already been incorporated into a design sent to manufacturing.
- OpenAI’s GPT‑6.2 Astra model delivered the best results on open RISC‑V cores for the Jalapeño inference chip.
- Anthropic’s Mythos model reduced prompt‑processing time on AMD MI355X GPUs from six‑twelve months to three days.

## Why it matters

Consumers could see graphics cards that run cooler and use less power, which translates to lower electricity bills and quieter laptops. Data‑center operators may be able to buy fewer chips for the same workload, cutting capital costs. The exact price impact remains unclear until the new designs reach the market.

## Sources & references

1. [OpenAI says its models delivered a year's worth of chip optimization on top of AMD's own engineering](https://the-decoder.com/openai-says-its-models-delivered-a-years-worth-of-chip-optimization-on-top-of-amds-own-engineering/) – The Decoder, 2026-11-18

Last updated: 2026-11-18
