Different Accuracy when inference model on PC and Orin Xavier

Hi guys,
I am trying to run a model pytorch on PC and Xavier ORin ( Jetpack 5.0).
My observation: the results of the model run on 2 devices are entirely different. I wonder if there is something special about hardware setting up here? I used the same code, the same model and even tried on some different models.
Could you please give me some ideas about these strange behaviours?
Thank you so much.
Update: the reason I saw that if you run your model on CPU mode of ORin xavier (cpu_orin) the prediction is NaN and It works if you run model on Gpu of Orin Xavier.
So my updated question is: Could you give me an idea of how to explain the difference between CPU and GPU of Xavier Orin.