Scale-out
Connecting many nodes into a looser cluster fabric (InfiniBand or Ethernet) to scale beyond one coherent domain.
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
~806–928 GB/s (2–32 GiB; sweep average ~872 GB/s)GB200 NVL72 in-domain all-reduce busbw across 2–32 GiB buffers (one provider's 72-rank run); scale-out gate set separately
~18× GB200 / ~9× GB300NVLink 5 per-GPU scale-up (~900 GB/s per direction) vs one scale-out NIC per GPU: ~18× GB200/ConnectX-7/400G, ~9× GB300/ConnectX-8/800G