OpenAI says its models improved AMD GPU designs by up to 14%
OpenAI reported that model-written circuit rewrites gave up to 14% better power, performance and area on AMD’s flagship GPU, and that some changes went into a design sent for manufacturing.
Source published · In the Daily Brief of 20 November 2026
Filed under AI industry, Agents
OpenAI said on 18 November that its models rewrote SystemVerilog code describing chip circuits and achieved gains of up to 14% in power, performance and area (PPA) against heavily optimised baselines for AMD’s flagship GPU. Synopsys tools were used to check that each rewrite preserved the block’s behaviour. AMD included some of the changes in a design sent for manufacturing and added others to production files for future versions. OpenAI described the GPU work as roughly a year of progress done in under five months.
On its own inference chip, Jalapeño, OpenAI estimates that model-proposed changes cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit. One multiplier redesign used about 15% less area and cut latency from two clock cycles to one, and the gains were confirmed in silicon.
Unite.AI noted a caveat in the report: some specification-based scores may be inflated because the models left out features, such as performance counters, that barely affect throughput.