OpenAI AI Model Cuts AMD Chip Design Cycle From One Year to Under Five Months

OpenAI's large model helps AMD cut chip design time from one year to under five months, achieving up to 14% PPA improvement on flagship GPUs.

AMD AMD
AMD AMD

OpenAI + use Gen AI in hardware design process, which is a major shift for how circuit engineers optimize. This partnership enables shortening an entire year worth of engineering cycle down to under 5 months. Buyers and watchers take note: that means a faster rollout of next generation silicon. It's about an advanced GPU and CPU design that will roll out faster.

AI optimization accelerates next-generation silicon rollout

The use of large language model is a real world application. For a very difficult physical engineering problem.

Here's the gist: OpenAI's large model is re-writing parts of AMD's SystemVerilog RTL codes on their flagship GPU. AMD is using Synopsys EDA tool chain to validate the function of AI re-written changes.

The whole thing is part of AMD's chip design pipeline. AMD can quickly iterate without re-writing codes manually for every variant. The problem it's aimed to solve is specifically for the verification and optimization of circuits for high end computing chips.

AMD says this AI-driven engineering approach has led to up to 14% of PPA (Power / Performance / Area) efficiency improvement. Also, they validated PPA single digit % improvement on some critical modules on next gen flagship CPU.

PPA means energy efficiency improvement or smaller die size. We don't know the detailed breakdown of power vs performance.

The optimized chips are already being part of tape-outs (chips going to be manufactured). OpenAI + AMD hasn't said which consumer/DC products are using the AI-optimized designs in the near future.

The up to 14% PPA improvement is power/performance/area improvement altogether, not a pure 14% increase in raw GPU speed. OpenAI said that their large model has been deeply involved in the design process. It has the effect of shortening 1 year of engineering to just 5 months. It shows how AI can shorten development cycle (time/cost) for semiconductor.

Comments (2)

    1. It is a composite metric for power, performance, and area. It does not mean a flat 14% FPS boost. It means the chip runs cooler or smaller for the same speed.

Add a comment

Keep it civil, relevant, and useful to other readers.