Editorial illustration for OpenAI says its models boosted AMD chip design, accelerated custom chip development
OpenAI Models Cut AMD Chip Design Time in Half
OpenAI says its models boosted AMD chip design, accelerated custom chip development
OpenAI says it shaved months off AMD's chip design process by putting its models to work on the company's flagship GPU IP. The project ran less than five months and produced up to a 14 percent improvement in power, performance, and chip area, a metric engineers call PPA. OpenAI frames that as roughly a year of extra optimization layered on top of work AMD's own engineers had already refined for years. Some of those changes have already made it into a design AMD sent off for manufacturing, according to OpenAI.
The same approach fed into Jalapeño, OpenAI's first custom inference chip, and into testing against open RISC-V processor cores, where the company's GPT-6.2 Astra model posted the strongest results yet across its model lineup. OpenAI is also pairing up with Synopsys on a model called GPT-Synopsys, built to run on Synopsys's EDA tools and aimed at making AI-assisted chip design usable beyond specialist teams.
Anthropic is pursuing a similar line of work with its own hardware partners, applying its Mythos model to a task on AMD MI355X GPUs that normally takes engineers months to finish.
Working with AMD, OpenAI achieved up to 14 percent PPA improvement (power, performance, and chip area) on AMD's flagship GPU IP in less than five months. According to OpenAI, that equals about a year's worth of progress on top of designs AMD's engineers had already optimized for years.
Why this matters
For anyone building hardware-adjacent AI tools, this is a concrete data point rather than a vague promise about "AI for chip design." A 14 percent PPA gain in under five months on AMD's flagship GPU IP, folded into a design already headed to manufacturing, is the kind of claim engineering teams can actually test against their own workflows. It also matters that OpenAI is using the same pipeline internally for Jalapeño, its own inference chip, and reporting that GPT-6.2 Astra outperformed prior model generations on RISC-V cores. That's OpenAI grading its own homework, so we'd want to see AMD or independent engineers quantify the gains before treating "a year's worth of optimization" as settled fact.
Still, if model-assisted EDA work is good enough that a company like AMD ships it, that's a signal worth watching for anyone in custom silicon, from startups designing accelerators to researchers studying where LLMs add real value versus where they just speed up busywork. The next thing to watch: whether AMD or other chipmakers publicly confirm specific PPA numbers on shipped products.
Common Questions Answered
What specific improvements did OpenAI's models achieve for AMD's flagship GPU IP?
OpenAI's models delivered up to a 14 percent improvement in PPA (power, performance, and chip area) on AMD's flagship GPU IP in less than five months. According to OpenAI, this level of optimization is equivalent to approximately one year of additional refinement work on top of the designs AMD's engineers had already optimized for years.
How long did the OpenAI and AMD chip design project take to complete?
The project ran for less than five months from start to finish. Despite this relatively short timeframe, the collaboration produced significant optimization results that have already been incorporated into designs AMD sent for manufacturing.
What does PPA stand for in the context of AMD's chip optimization?
PPA stands for power, performance, and chip area, which are key metrics that engineers use to evaluate chip design efficiency. The 14 percent improvement in PPA means OpenAI's models enhanced all three of these critical aspects of AMD's GPU design simultaneously.
Why is OpenAI's collaboration with AMD significant for hardware-adjacent AI tools?
This collaboration provides concrete, measurable evidence of AI's capability in chip design rather than just theoretical promises. The 14 percent PPA gain achieved in under five months on a flagship GPU design that's already heading to manufacturing gives engineering teams a real data point they can evaluate against their own workflows and use cases.
Is OpenAI using the same optimization pipeline for its own chip development?
Yes, OpenAI is using the same pipeline internally for Jalapeño, its own inference chip, demonstrating confidence in the methodology developed through the AMD collaboration. This indicates that the techniques proven effective for AMD's GPU optimization are being applied to OpenAI's proprietary hardware projects.
Further Reading
- Advancing chip design with OpenAI models - OpenAI
- OpenAI says its models delivered a year's worth of chip optimization on top of AMD's own engineering - The Decoder
- OpenAI says its AI found 14% gains in AMD's flagship GPU design - The Next Web
- OpenAI models deliver up to 14% PPA improvement in AMD chip design - AI Radar Haber