OpenAI and Broadcom presented Jalapeño as an inference accelerator built to run OpenAI's large language models and other inference workloads, with the companies reporting engineering samples are operating at target frequency and power.
OpenAI, the frontier model lab, framed the chip as the first in a planned family of AI accelerators that will let it control the full inference stack, chip architecture, kernels, memory systems, networking, scheduling, deployment systems and product experience, to make models faster, more reliable and more affordable.
"Jalapeño was co-developed from initial design to manufacturing tape-out in just nine months," the companies said in a press release.
OpenAI said the design process used its models to accelerate parts of chip development, Broadcom handled implementation and manufacturing, and early testing shows the chip will deliver performance per watt substantially better than current state-of-the-art, with technical details to be released in a performance report in the coming months while other cloud and AI giants continue to develop their own silicon.
