Core Viewpoint - NVIDIA is leveraging a new midplane design in its next-generation AI server, the Rubin NVL576, to enhance computing density and efficiency in the competitive AI server market [1][6]. Group 1: Design Innovations - The NVIDIA Rubin NVL576, codenamed "Kyber," features a new rack design that removes the cable box and relocates fans and power supplies outside the rack to increase computing density [1]. - The new midplane design includes 18 columns and 4 rows of connectors, allowing for improved internal connectivity [5]. - Liquid cooling is employed throughout the rack, enabling the midplane design without the need for airflow considerations, which traditionally limited density [6]. Group 2: Performance Enhancements - The new design allows for a significant increase in copper interconnect coverage, which can save substantial power and enhance optical reliability [8]. - The updated backplane design facilitates more silicon connections through copper interconnects within the same rack, which is essential for the next generation of AI clusters [8].
NVIDIA Rubin NVL576 Kyber 中板,非常巨大