NVIDIA Jetson Nano 2GB Developer Kit (945-13541-0000-000)

I use this to process deep learning models at the edge. It worked out of the box with an iPod power supply. I quickly discovered that a fan was a key requirement to avoid throttling the CPU speeds as the temperature rose. I chose the Noctua NF-A 4x20 PWM fan, silent and effective. The heat sink in the Nano used to be too hot to touch, but it runs cool and fast even with the Nvidia bubble demo app. (cd /usr/src/nvidia/graphics_demos/bubble/ & make clean & make & ./x11/bubble). No throttle. Also love the docker support. I use this one: (sudo docker pull nvcr.io/nvidia/l4t-ml:r32.5.0-py3 &sudo docker run -it --rm --volume ~/nvdli-data:/nvdli-nano/data --runtime nvidia --network host nvcr.io/nvidia/l4t-ml:r32.5.0-py3 )it has all I need and is preset to use the 100 cores to maximize performance. Truly amazing little computer.

This is not a competitor to the Raspberry Pi. Don't expect it to be friendly to the novice user. If you are not an embedded engineer, this is probably not the board for you. If you are, however, it offers a lot of FPU per unit dollar, and glsl/cuda just works, right out of the box, without having to do any of the weird jury-rigged routines you need to get that kind of functionality on Raspberry.If you need to do serious graphics or AI work on a small board, buy this. If you just need a small board with embedded linux, and lots of aftermarket support, buy a Raspberry.

The 2 GB version may not be of any value for machine learning but it is great for other things. It works a direct play plex server. It can't transcode over CPU or GPU on plex, but can transcode on GPU with a very specific and modified FFMPEG version. It can also play some java edition minecraft. Not at great settings but still something. As a traditional desktop, it is not great. Most developer tools are not made ARM friendly so you'd have to make stuff from scratch. Still, better than PI and very fun. Does require a power cable and a fan or it will burn up...

This unit feels much quicker than the Raspberry Pi 4. There is no fan so the huge heat sink gets warm to the touch which means it’s doing it’s job and it’s totally silent!There no case so be careful not to have any metal objects or other conductive material where you place it. I plan to 3D print a case for it.I’m just getting stared with it an have the docker container loaded for the demo programs - will be a great learning tool.

I initially bought this to pair with a server PC that would be doing the actual heavy lifting for training. It turns out that 2GB is not enough for the OS + Pytorch when using CUDA which makes it effectively useless for development because compiling the model from Pytorch -> ONNX -> TensorRT is unfeasibly slow and hits the swap space.If you want to compile a model with TensorRT though, it runs like a dream so for deployment this is probably plenty. Just a fair warning, you likely want the 4GB one insteadAnd as others have said, get a good SD card because it's gonna be hit a lot

The 2GB model is different than the developer kit version. This version DOES NOT have the slot under the processor for the wifi card. It comes with a USB WiFi dongle and short USB extension cable. If you are watching Paul McWhorters YouTube series on the Jetson Nano, ignore his reference to the WiFi kit, power supply & case. The 2GB board has some differences & may not work with some A01 & B01 series cases. Also you need a USB-C power supply as there is no coaxial type jack as there is on the developer series. Otherwise it's a great board with exceptional on-line support unlike a certain competitors AI product. The online support make the Nvidia product superior.

This version of Jetson Nano need more access to swap file, because of 2GB RAM. This mean more read/write on SD card. Must choice "endurance" class SD card.If you are going to work on image processing, it is a good idea to install a fan in the CPU cooler. But fan connector is not populate on the board. Need a little bit soldering work. If you don't want to soldering works just buy 2 wire simple 5V DC fan and connect to 40 pin headers power pins (pin4 is +5VDC, pin 6 is GND).There no case so be careful not to have any metal objects or other conductive material where you place it.