SpaceX develops its own ‘Kessel’ rack-scale system featuring 72 Nvidia Rubin GPUs for data centers in space, report claims
SpaceX is reportedly developing a 'Kessel' rack-scale system design with 72 Nvidia Rubin GPUs that relies on a massive 84-layer PCB midplane.
SpaceX and xAI are crafting a custom Nvidia Vera Rubin NVL72 rack-scale AI system dubbed Kessel, claims SemiAnalysis. The system replaces traditional NVLink cable assemblies with an 84-layer printed circuit board (PCB) interconnect, retaining 36 Vera CPUs, 72 Rubin GPUs, and nine NVLink switch trays. SpaceX will procure certain components from third-party suppliers, potentially gaining greater control over its supply chain.
The Kessel design uses a PCB midplane similar to Nvidia's Kyber design, boasting an 84-layer PCB with ultra-low-loss M10 copper-clad laminates, enabling simplified system assembly and independent component procurement. However, the complex midplane could pose manufacturing challenges and increase costs. Kessel diverges from Nvidia's conventional system-level procurement model, as SpaceX could source hardware directly from Nvidia and other vendors, optimizing the system for orbital deployments and potentially reducing costs.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
