nvidia-smi collectors were deliberately SKIPPED.

NVSMI_STATE was 'error', i.e. the driver is not answering. Every
nvidia-smi invocation against a wedged driver blocks in uninterruptible (D)
state, cannot be killed, and holds a reference on the nvidia module for the
rest of this boot -- which is precisely what prevents the module from being
unloaded and the box from shutting down cleanly.

Collecting them would have made the fault worse, so they were not run.
See dmesg_nvidia.txt and xid_history.txt for what the driver reported instead.