How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

OpenAI’s Jalapeño AI accelerator delivers 13.4 petaflops of 4-bit compute and 15.4 terabytes per second of memory bandwidth, reducing inference latency by 3.6 times compared to Nvidia’s GB300. A team of fewer than 100 engineers utilized LLMs to complete the chip's design from architecture to silicon in under 20 months, partnering with Broadcom for physical implementation.