OpenAI says its models found up to 14% improvement in power, performance and area (PPA) trade-offs on AMD’s heavily optimized flagship GPU circuitry in less than five months. According to its November 18 announcement, some changes are already included in a design AMD sent for manufacturing. Source
“Heavily optimized,” apparently, didn’t mean out of ideas. But this isn’t a 14%-faster-GPU announcement: the claim concerns circuit-design improvements, and OpenAI doesn’t disclose how it calculated the headline percentage or divided the gains among power, performance and area. Source
What the models actually rewrote
The work involved rewriting SystemVerilog RTL, the code that describes a chip’s circuits. OpenAI says the model used Synopsys tools to check that each rewrite preserved the affected block’s behavior and to evaluate the power, performance and area trade-offs. That describes circuit optimization within an engineering workflow, not an AI independently delivering an entire production GPU. Source
OpenAI calls the result “roughly a year of progress” achieved in less than five months. That’s its assessment of the design gains, not a measure of time cut from AMD’s overall chip-development schedule. Source
Beyond the GPU work, OpenAI says AMD verified single-digit percentage PPA gains across key blocks of an upcoming next-generation flagship CPU using model-written designs. The 14% maximum does not extend to that CPU claim. Source
A manufacturing handoff, not a finished-chip benchmark
OpenAI says AMD incorporated some GPU changes into a recent design submitted for manufacturing and added others to production design files for future versions. The announcement does not identify the GPU, specify the submission date or report measurements from finished AMD silicon incorporating those changes. Source
OpenAI says model-proposed optimizations for Jalapeño, its first custom inference chip, were proven during silicon bring-up. Its team estimates that the changes reduced standard-cell area in Jalapeño’s matrix unit by about 10%. Those silicon receipts belong to Jalapeño, not AMD. Source
Synopsys’s role extends beyond checking circuit rewrites. OpenAI says the companies are partnering to train GPT-Synopsys, a new model intended to use Synopsys’s electronic design automation tools as expert chip designers do. The post doesn’t detail public availability or pricing. Source
Comments