Review on NVIDIA Jetson Xavier Developer 32GB by Emanuel Frison

First, I have several Jetson products. I have a couple of Nanos and they are great for the price, but since I'm using Python and Tensorflow it takes up a lot of RAM and runs a lot of processes (duplicates RAM usage). I'm not sure what's going on, but it's a Tensorflow issue, not an Nvidia issue. I bought this device to run many AI models in parallel and due to memory issues, I needed a lot of RAM. If you're short on memory and don't train models with a device, choose one of the cheaper Jetsons. Nano is a great value and AI is very good at predicting it. I'm using a Tornado server and evaluating three networks in less than 50ms. I'm not a typical use case for this device. I wouldn't say it's a bad buy but if you don't need RAM go for lower end models which are amazing for the price. Train your models to 2070 or higher (if you need tensor core speed). UPDATE: I'm now running a Docker swarm with Nanos and Xavier. I have RAM under control and my conclusions are: For a simple AI prediction I get about twice the performance of the Nano and twice the RAM usage (because TF executes processes by number of cores, which is 8 vs. 4 on the Nano). For my small neural networks, that's overkill. I still think it's an amazing device, I don't need the power right now. This device gets very hot. Also, it will NOT turn on automatically when you have an electrical event. There is a way to jumper multiple pins so it turns on automatically, but I haven't done that yet. Whether you train networks or make AI for videos, this Jetson is for you. If you make predictions less frequently and don't need RAM, get NX.