2.1 From CPU to GPU — what your fabric inherited
Why CPUs can't do AI (the four walls), and how the shift from CPU workloads to GPU workloads turned your network from a busy restaurant into a synchronized assembly plant.
2.2 Inside GPU Anatomy — Hardware
The hardware view of a DGX-class AI server — vendor landscape, the dual-tray architecture, and every component in the chassis. Pair with the CLI page for the shell view.
2.3 Inside GPU Anatomy — CLI Version
The DGX H100 server walked component by component from the shell — nvidia-smi, NVLink mesh, ConnectX-7 RDMA NICs, PCIe + NUMA, DIMMs and storage, BMC + power, and the pre-flight checks every node must pass before it joins a training cluster.