OpenAI's models optimized AMD's flagship GPU IP, achieving up to 14 percent PPA improvement in less than five months.
TL;DR
OpenAI's AI models helped AMD improve its flagship GPU IP by up to 14 percent in power, performance, and chip area. This progress, achieved in under five months, is equivalent to about a year of human engineering work. Some of these AI-generated changes are already in manufacturing.
Key facts
- Maker: OpenAI
- What's new: AI models optimized chip design for AMD's flagship GPU IP
- Benchmark: Up to 14 percent PPA improvement on AMD's flagship GPU IP
- Context: Equals about a year's worth of progress on top of existing designs
- Other use: Helped with Jalapeño, OpenAI's first custom inference chip
What is it?
This is a project where OpenAI used its AI models to optimize the design of computer chips. The focus was on improving power, performance, and chip area, known as PPA. The work was done in collaboration with AMD on their flagship GPU IP.
How does it work?
OpenAI's models analyzed and suggested improvements for chip designs. This process delivered significant optimizations in a short timeframe. The same method was also applied to OpenAI's own custom inference chip, Jalapeño, and tested with open RISC-V processor cores.
Why does it matter?
This approach shows that AI can accelerate complex engineering tasks like chip design. It can achieve improvements that would take human engineers much longer. This could lead to faster development cycles and more efficient hardware.
Who is it for?
Chip designers, hardware manufacturers, and AI researchers interested in applying AI to engineering problems.
Try it
Not available yet