GPU interconnect fabric

The coherent high-bandwidth link (NVLink / ConnectX-class) that binds multiple GPU nodes into one memory-and-compute fabric for multi-node inference.

grounded in: spark GB10 telemetry commits (afb4e9e '2-node GB10 telemetry refresh') — a 2-node GB10 setup serving inference across a coherent interconnect

Connected concepts

Distributed inference, Unified memory inference, 2-node NVIDIA GB10 (DGX Spark), vLLM / PagedAttention