← The AI Systems Map
L6 · PowerNetworking⬦ ChokepointUSA

NVLink / NVSwitch

NVIDIA's GPU-to-GPU scale-up interconnect

NVLink is NVIDIA's proprietary high-bandwidth interconnect that ties GPUs together into a single scale-up domain; NVSwitch extends it across a chassis. In the GB200 NVL72 rack, fifth-gen NVLink delivers 1.8TB/s per GPU and unifies 72 Blackwell GPUs into one coherent NVLink domain (~130TB/s aggregate). NVLink Fusion (2025) opens the fabric to third-party CPUs/accelerators, deepening NVIDIA's platform lock-in as the scale-up standard.

5th-gen bandwidth

1.8TB/s per GPU

NVL72 domain

72 GPUs, ~130TB/s aggregate

NVLink Fusion

2025 opening to 3rd-party silicon

How it fits the stack

NVLink / NVSwitch with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.

depends onusescompetes withcompetes withcompetes withdesignsNVLink / NVSwitchPowerNCCL / RCCLcollectiveschokepointNvidiachokepointCo-packaged optics &silicon photonicsUALink ConsortiumUltra Ethernet /UALink consortia
NVLink / NVSwitchFeeds ↓Related

NVLink / NVSwitch in the AI stack. NVLink / NVSwitch with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.

Graph data (text) — 6 entities, 6 relationships
  • NCCL / RCCL collectivesdepends onNVLink / NVSwitch
  • NvidiausesNVLink / NVSwitch
  • NVLink / NVSwitchcompetes withCo-packaged optics & silicon photonics
  • NVLink / NVSwitchcompetes withUALink Consortium
  • NVLink / NVSwitchcompetes withUltra Ethernet / UALink consortia
  • NVLink / NVSwitchdesignsNvidia