**Priority:** High **Context.** NVLink fabric is currently faked only at the *identity* level (clique/cluster UUID via ComputeDomain #337/#342). **Gap.** No `nvidia-fabricmanager` daemon, no NVSwitch device modeling, no NVLink per-link state/counters, no `nvidia-smi nvlink` / `nvidia-smi topo -m` output. **Proposed scope.** - NVML NVLink APIs (link state, capabilities, BW/error counters) backed by profiles. - `nvidia-smi topo -m` matrix derived from the PCIe/NVLink topology already rendered (#353). - Optional fake `nvidia-fabricmanager` readiness analogous to fake IMEX. - E2E: assert `nvidia-smi topo -m` shows NVLink links and nvlink counters grow. **Why.** Closes the data-plane half of the NVLink story whose identity half already shipped.
Priority: High
Context. NVLink fabric is currently faked only at the identity level (clique/cluster UUID via ComputeDomain #337/#342).
Gap. No
nvidia-fabricmanagerdaemon, no NVSwitch device modeling, no NVLink per-link state/counters, nonvidia-smi nvlink/nvidia-smi topo -moutput.Proposed scope.
nvidia-smi topo -mmatrix derived from the PCIe/NVLink topology already rendered (feat(nvml-mock): add PCIe topology to profiles and render mock sysfs tree #353).nvidia-fabricmanagerreadiness analogous to fake IMEX.nvidia-smi topo -mshows NVLink links and nvlink counters grow.Why. Closes the data-plane half of the NVLink story whose identity half already shipped.