rbtfl.

tech / semiconductors

Par parti pris · 1 takes across the edition

Reports OpenAI's first in-house inference processor, Jalapeño, designed with Broadcom, claiming significantly better performance-per-watt than current state-of-the-art parts, still in testing with no deployment date, and framed as a bid to cut inference cost and reduce Nvidia dependence.

“OpenAI says Jalapeño shows significantly better performance-per-watt than current state-of-the-art alternatives, though it remains under evaluation.”