adding a script to catch a generic nvidia-smi error

this should catch the issue on the m3a nodes and similar in the future

tested on m3p006 ( good case ) and m3a101 ( bad case )