Scale-up
Tightly coupling GPUs into one coherent high-bandwidth domain (NVLink class) that acts like a single large accelerator.
Current numbers
~18x (GB200, 400G) / ~9x (GB300, 800G) / ~7.5x (Vera Rubin, 1.6T)scale-up versus back-end scale-out peak bandwidth per GPU: roughly 8-18x per direction by generation
72 GPUs / 130 TB/sGB200 NVL72 scale-up domain; topology membership does not require whole-domain allocation
HGX 8; NVL72 72; Rubin Ultra NVL576 576/8 racks; Kyber NVL144 144/rack, NVL1152 1,152/8 racksscale-up (NVLink) domain size: HGX node, NVL72 rack, announced Rubin Ultra and Kyber