OpenAI processor surpasses Nvidia in efficiency

OpenAI has revealed the performance results of its proprietary AI processor, Jalapeño, which outperformed powerful processing systems like Nvidia Blackwell in response speed and energy efficiency. The processor achieved a performance that is between 1.5 and 1.9 times higher per watt at maximum operating speeds, along with reducing response times by 1.7 to 3.6 times compared to the top competitors, based on independent tests involving various models. Developed in collaboration with Broadcom, the processor aims to enhance the efficiency of running AI models during inference, lower energy costs, and increase response speed—especially in intelligent agent applications that require executing long sequences of steps at high speed. OpenAI plans to incorporate it into their computing infrastructure in limited quantities by the end of 2026, with ongoing development of the second and third generations of processors to achieve greater performance, even as they continue to rely on partners like Nvidia.

Size: 120%Text Size

Notice: This Is an AI-Generated Summary

Display The Full Article