The companies attributed this speed to a deep software-hardware co-development process that actively used OpenAI’s own models ...
Named Jalapeño, the new processor was designed specifically for the unique needs of OpenAI's inference systems.