OpenAI Optimizes AMD Chip Designs Using AI Models
OpenAI used AI models to improve AMD GPU and CPU performance by up to 14%, condensing a year of design work into five months.
OpenAI partnered with Advanced Micro Devices to optimize the design of upcoming flagship GPUs and CPUs using undisclosed AI models. By rewriting SystemVerilog register-transfer level code, the models achieved up to 14% improvements in power, performance, and area (PPA) for GPU circuitry and single-digit PPA gains for processors. OpenAI stated that this collaboration achieved "roughly a year of progress achieved in less than 5 months."
AMD has already integrated several of these AI-generated optimizations into designs submitted for manufacturing and production files for future versions. Following the announcement, AMD shares rose 5.3% on November 18 to close at $530.14.
Beyond the AMD partnership, OpenAI applied these models to its own custom AI inference accelerator, Jalapeño, developed in collaboration with Broadcom Inc. The company reported that model-proposed changes reduced the standard-cell area in the chip's matrix unit by approximately 10%, a result proven in silicon. Testing on open-source RISC-V cores indicated that the GPT-6.2 Astra model was the most effective for automated circuit optimization.
To expand these capabilities, OpenAI is collaborating with Synopsys to train GPT-Synopsys, a specialized model designed to operate professional electronic design automation and verification tools. Synopsys tools were also used to verify that the AI-rewritten circuits preserved their original behavior.