This is an exciting time for small factorial calculations. As if the Raspberry Pi lacked a versatile machine, more powerful boards continue to emerge that are capable of incredible feats.
Nvidia’s Jetson Nano is a recent addition to its range of super-powerful machine learning boards. What makes it special? Is it worth buying? What is Nvidia Jetson Nano?
What is Nvidia Jetson Nano?
The Jetson Nano is a Raspberry Pi-sized single board computer (SBC) designed for artificial intelligence and machine learning. Seemingly a direct competitor to Google’s Coral Dev board, it is the third in the Jetson family along with the already available TX2 and AGX Xavier development boards.
Nvidia is using its prowess to process graphics information on these small computers, using parallel neural networks to process multiple videos and sensors at the same time.
While all three Jetson boards aim to be accessible to everyone, the Nano is aimed at hobbyists and professional developers alike. The development kit consists of two parts: the main board for connecting and the System On Module (SOM) for real processors.
What is a system on a module?
A system-on-module refers to any development board that has all system-critical parts in a plug-in module. The Nano is equipped with a 260-pin edge connector to attach it to a development board.
Once development is complete, the SOM can be removed and added to the embedded system with user inputs, and the new SOM is plugged into the base board for further development.
If this all sounds a bit familiar, it is!
This is the same setup as Google Coral Dev Board. which is similar in size, and is also designed for machine learning embedded machines for hobbyists and professionals!
What are the characteristics of the Jetson Nano?
Nvidia has packed a lot into the Jetson Nano:
- Processor: Quad-core ARM® Cortex-A57 MPCore
- GPU: Nvidia Maxwell™ architecture with 128 Nvidia CUDA cores
- RAM: 4 GB, 64-bit LPDDR4
- Memory: 16GB eMMC 5.1 Flash
- Video: 4k@30fps encode, 4k@60fps decode
- Camera: 12 lanes (3×4 or 4×2) MIPI CSI-2 DPHY 1.1 (1.5Gbps)
- Connectivity: Gigabit Ethernet
- Display: HDMI 2.0 or DP1.2 | eDP 1.4 | DSI (1×2) 2 simultaneously
- PCIE / USB: 1 x1 / 2/4 PCIE, 1x USB 3.0, 3x USB 2.0
- I/O: 1x SDIO / 2x SPI / 6x I2C / 2x I2S / GPIO
- Dimensions: 69.6mm x 45mm
- USB: 4x USB 3.0, USB 2.0 Micro-B
- Camera: 1 lane MIPI CSI-2 DPHY (compatible with Raspberry Pi camera)
- LAN: Gigabit Ethernet, M.2 Key E
- Storage: microSD slot
- Display: HDMI 2.0 and eDP 1.4
- Other I/Os: GPIO, I2C, I2S, SPI, UART
What can it do?
To no one’s surprise, Nvidia has designed a board well-suited for visual tasks. Object recognition is key here, and the Visionworks SDK has many potential applications in this area.
Instead of using a separate processor for machine learning tasks, Jetson Nano uses a Maxwell GPU with 128 CUDA cores to do the heavy lifting.
The Jetson Inference project provides demonstrations of a pretrained neural network that provides high-performance multi-object recognition in a variety of environments. Feature tracking, image stabilization, motion prediction, and multi-source processing are all available in the available demo packages.
Perhaps most impressive is the DeepStream technology shown in the video above. Running real-time analytics on eight simultaneous 1080p streams at 30fps on a small single board computer is incredible and shows the potential power of the Nano hardware.
What will it be used for?
Given its video analytics prowess and small form factor, the Jetson Nano will almost certainly prove itself in robotics and autonomous vehicles. Many of the demos show these applications in action.
Given its power and size, it is also likely to work in embedded systems that rely on face and object recognition.
For lovers like us? It seems to be the perfect combination of powerful machine learning capabilities with a factor familiar to anyone who has messed around with the Raspberry Pi. While you can use machine learning frameworks like TensorFlow on Raspberry Pi, the Jetson Nano is much more suitable for this task.
What else can the Jetson Nano do?
The Jetson Nano runs Ubuntu, although a special OS image is available from Nvidia with platform-specific software. While the main focus of the board is machine learning, it’s Nvidia, so you’d expect some graphical magic to happen as well.
You won’t be disappointed. Demos showing particle systems, real-time fractal rendering, and a host of visual effects have until recently been found on flagship desktop graphics cards.
Considering its video encoding is rated at 4k at 30fps and decoded at 60fps, it’s safe to assume the Nano will be ideal for video applications as well.
Jetson Nano vs Coral Dev Board: which is better?
At this stage, it’s hard to say which is the best board between Google Coral Dev and Jetson Nano.
The Google TensorFlow neural network is the dominant force in the field of machine learning. It follows that the native Google Edge TPU co-processor might work better for TensorFlow Lite apps.
On the other hand, Nvidia has already shown off an impressive array of machine learning-based demos for the Jetson Nano. This, along with impressive graphics, makes the Nano a real contender.
How much does the Jetson Nano cost?
Price is another aspect we haven’t covered yet. The Google Coral Dev board is priced at $149.99, while the Jetson Nano is only $99. If the Coral Dev board doesn’t bring something unique, hobbyists and small developers might find the extra $50 hard to justify.
There is currently no SOM-only pricing for both boards, but I think for most hobby builders this won’t be that big of a deal. From a commercial standpoint, the difference between performance and price will be critical for the Jetson Nano and the Coral Dev board.
The Jetson Nano can be purchased from Nvidia directly from third party retailers.
Buy : Jetson Nano directly from Nvidia