OpenAI deploys self-developed AI chip Jalapeño ASIC
OpenAI has internally deployed its self-developed Jalapeño ASIC AI chip. The chip is used in conjunction with AMD EPYC Turin processors.
AI company OpenAI has internally deployed its self-designed Jalapeño ASIC chip, according to technology publication Tom's Hardware. The chip is being used with servers equipped with AMD EPYC Turin processors featuring 1.5 terabytes of memory.
OpenAI's VP and Head of Hardware, Richard Ho, stated that the choice of the Turin platform was made for practical reasons. He noted that Nvidia's new Vera CPU is "slightly behind" in terms of maturity. Ho emphasized that AMD's chips offer robust performance that meets the full requirements of the Jalapeño project, and that partner familiarity with the platform aids in rapid deployment.
The Jalapeño project's design principles focused on "de-risking" and "rapid implementation," Ho explained. The company aimed for aggressive performance and cost control while adhering to existing engineering frameworks to ensure project manageability.
The Jalapeño ASIC is currently for internal use only at OpenAI, with no current plans for external sales. However, Ho indicated that broader deployment could be considered in the future if resources permit.