OpenAI models helped AMD improve its flagship GPU design by up to 14% in PPA (power, performance and area). Some changes have already been sent for manufacturing.
What happened?
OpenAI reported that its models rewrote parts of AMD's SystemVerilog RTL code, finding optimizations equivalent to about a year of progress in less than five months. AMD verified single-digit gains in blocks of its next flagship CPU.
On OpenAI's internal Jalapeño chip, the changes reduced standard-cell area by about 13% in the vector unit and 10% in the matrix unit, confirmed in silicon.
Why does this matter?
This demonstrates AI's potential to accelerate complex chip design. OpenAI president Greg Brockman described the results as unprecedented at the frontier of chip design.
The collaboration occurs in the context of larger agreements between the companies for Instinct GPUs.
What changes in practice?
PPA improvements could lead to more efficient GPUs in future AMD generations. There are no details yet on when these chips will reach the market.
By Grok