Scale-up
Tightly coupling GPUs into one coherent high-bandwidth domain (NVLink class) that acts like a single large accelerator.
Current numbers
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 Kyber
~18× GB200 / ~9× GB300NVLink 5 per-GPU scale-up (~900 GB/s per direction) against one per-GPU scale-out NIC — ~18× on GB200/CX-7/400G, ~9× on GB300/CX-8/800G; keep collectives in scale-up
512 XPUssingle-hop all-to-all scale-up domain on one Broadcom Tomahawk 6 (102.4 Tbps, SUE/ESUN)