I also get the same message but let me describe my context. I own the Jetson Nano. I am doing a show and tell by using a pre-trained model for image recognition (Works great with no Throttle message), I trained a Model on my Nano (It took an hour to complete but with success and no Throttle message), I used the Transfer Learning Technique to train a new model (Again no issues - no messages). This was using the NVidia Docker ML image 4.4.1. This little machine is powerful .This excitement convinced us to purchase the Xavier NX. Installed the newer NVidia Docker image 4.5.1 - Using our same models and Notebook code, On the pre-trained model we get the Throttle message, on the Training our own model we do not get the message but model accuracy was cut from 92% to 54%, on the Transfer learning model, we do not get a message the model accuracy is cut from 96% to 52%. We are running the NX on 15W 6Core. We finished faster but the accuracy is horrible compared to the 90% plus found on the Nano. We are now trying determine what may be happening here. Thoughts? And Thanks for any input!