OpenAI Reports Model-Driven Gains in AMD Flagship Chip Designs

Unite.AI
Read full story· 5 min
OpenAI reported that its AI models improved AMD flagship GPU chip designs by up to 14% in power, performance, and area through SystemVerilog RTL rewrites, verified with Synopsys tools. These optimizations, applied to the Jalapeño inference chip and tested across RISC-V cores, significantly reduced cell area and latency, accelerating chip design timelines.

What happened

On November 18, 2026, OpenAI announced that its AI models delivered up to 14% improvements in power, performance, and area for AMD's flagship GPU designs by rewriting SystemVerilog RTL code. These changes were verified using Synopsys tools and some were included in a design sent for manufacturing. OpenAI also applied its models to its own Jalapeño inference chip, achieving notable reductions in cell area and latency. The company tested multiple model generations on RISC-V cores to benchmark design optimization capabilities.

Key details

  • OpenAI's models achieved up to 14% PPA (power, performance, area) improvement on AMD GPU IP.
  • Jalapeño chip's warp engine area was reduced by about 13%, and matrix unit by 10%.
  • Matrix-unit multiplier redesign cut cell area by 15% and reduced latency from two to one clock cycle.
  • GPT-6.2 Astra scored highest in RTL optimization tests on PicoRV32 and Ibex RISC-V cores.
  • OpenAI and Synopsys have a multi-year agreement to develop GPT-Synopsys for EDA tool integration.

Why it matters

Chip designers at companies like AMD can now use AI models to accelerate hardware development, achieving in months what previously took a year. This enables faster time-to-market for new chips and reduces design costs. The integration with Synopsys tools means that these AI-driven optimizations can fit into existing industry workflows while maintaining security and verification standards for proprietary designs.


More in Chips & Compute