NVIDIA Jetson Nano It is a special development board. It can look like your own in many ways Raspberry Pi, or Arduino, but it is specifically designed for a specific type of project. And like these other development boards, it is also reasonably low priced and small in size compared to alternative equipment.
Specifically, NVIDIA's Jetson Nano is specifically targeting the development of artificial intelligence and artificial neural networks projects. A cheap way to start in this world, learn how these intelligent systems work, and create an infinite number of projects that you can imagine ...
Table of Contents
What is Jetson Nano?
NVIDIA Jetson Nano It is a development board, an SBC with which to create numerous projects based on neural networks, deep learning and AI. With it you can create very varied projects, from small intelligent IoT applications, to more complex robots, artificial vision systems and object recognition, devices that react intelligently by evaluating a series of sensor parameters, small autonomous vehicles, etc.
But all with a plate of a few dimensions, and with a price quite affordable compared to other professional systems with similar characteristics.
And if you wonder why should you have one of these NVIDIA Jetson Nano boards, you should keep in mind that these boards will allow you to create many projects while learning about a technology that is on the rise. There are more and more companies interested in people with knowledge of machine learning, AI, deep learning, and other similar disciplines, since it is a technology of the future.
NVIDIA Jetson Nano offers really impressive features for its size and price. It barely exceeds € 100, and with a few centimeters in size. Despite this, it can develop up to 472 performance GFLOPs, enough to run many AI algorithms very quickly and process multiple artificial neural networks simultaneously.
And it is not only impressive for these figures, but also for its low consumption. This board may have a consumption that is between 5 and 10W. Compared with similar systems it is certainly low, so you are facing a very efficient system. It has little to do with other machines that consume hundreds or thousands of watts ...
For more information, you can see this full details list:
- NVIDIA Maxwell GPU with 128 CUDA cores
- ARM Cortex-A57 QuadCore CPU
- 4GB RAM LPDDR4
- 16GB eMMC 5.1 flash storage
- 12-way camera connector (3 x 4 or 4 x 2) MIPI CSI-2 DPHY 1.1 (18 Gbps)
- Gigabit Ethernet network (RJ-45)
- HDMI 2.0 or DP 1.2 display connection | eDP 1.4 | DSI (1 x 2) 2 simultaneous
- Ports 1/2/4 PCIE, 1 USB 3.0, 3 USB 2.0
- Additional I / O: 1 SDIO / 2 SPI / 4 I2C / 2 I2S / GPIO
- 260-pin connector
- Size: 69,6mm x 45mm
- Consumption: 5-10w
- Linux OS with development kit
NVIDIA Jetson Family Products
NVIDIA has several of these products for AI development with artificial neuroanal networks. Some of the most prominent products are:
- Jetson Xavier NX: it is a SOM, that is, a System On Module, or a complete system integrated into a single module. Despite its appearance and size, it offers typical supercomputing powers, with up to 21 TOPs, that is, 21 Tera Operations per second. That's enough to run multiple artificial neural networks smoothly and simultaneously.
- Jetson AGX Xavier: another very powerful module in terms of computational density and efficiency and that has come after Jetson Nano, allowing the creation of new generations of intelligent machines.
- Jetson TX2: another alternative to Jetson Nano, and from the same family. It stands out for its enormous speed and energy efficiency. Especially designed for embedded AI applications, where size and consumption matter. In this case, it is based on the NVIDIA Pascal architecture, powered by 8GB of RAM and a bandwidth of up to 59,7GB / s.
Buy NVIDIA Jetson Nano
If you are willing to get started in the maker or DIY world with artificial neural network projects, you can buy this NVIDIA Jetson Nano board in specialty stores or on platforms like Amazon, where they are sold separately or with development kits to get started quickly with everything you need:
- NVIDIA Jetson Nano Basic
- NVIDIA Jetson Nano Kit with power adapter, 64GB microSD, USB
- Buy only the SOM module
Currently an NVIDIA Jetson Nano board has been launched with a reduced price of about $ 59 and to which they have also added WiFi. Great news, the only thing is that they have reduced the main memory to 2GB. If you want it you will have to wait, for now it is only in presale for partners ...
Alternatives to the NVIDIA Jetson Nano
If you are interested in machine learning, AI and artificial neural networks, you should know some alternatives to NVIDIA Jetson Nano, since it is not the only plate for these purposes. You can find some SBCs designed specifically for these projects like the following:
Google has developed a badge, Google coral, along with other accessories and modules needed to create AI projects. Among the articles belonging to this platform you have:
- NXP i.MX 8M CPU with Quad Core Cortex-A53 and Cortex-M4F
- GC7000 Lite Graphics GPU,
- Google Edge TPU coprocessor with up to 4 TOPS or 2 TOPS / w.
- Includes 1GB LPDDR4 RAM
- Storage of up to 8GB eMMC flash and the possibility of expanding it using microSD cards.
- It has WiFi connectivity, USB, Bluetooth, Ethernet, audio jack, HDMI, MIPI-DSI, and power over USB-C 5v.
Khadas VM3 It is another alternative for your AI projects, although it does not have some of the characteristics of the big ones, it is a fairly modest board that can be a good opportunity to start:
- CPU A311D x4 Cortex-A73 2.2Ghz and x2 Cortex-A53 at 1.8Ghz.
- With an NPU at 5 TOPS
- Up to 4GB of RAM
- 16-32GB eMMC Samsung
- MIPI-DIS, HDMI, WiFi, Ethernet, microSD, USB, PCIe connections, etc.
HiSilicon HiKey 970 (Huawei)
HiSilicon is the company under Huawei that manufactures the chips. Well, under this brand you will find another alternative to develop neural network projects such as HiKey 970, compatible with the Huawei SDK. In addition, it has some interesting features:
- ARM Kirin with Cortex A73 QuadCore + Cortex-A53 QuaCore
- GPU Mali G72 MP12
- Dedicated NPUs
- 6GB of LPDDR4
- 64GB flash memory
- WiFi, microSD, HDMI, USB, PCIe connections, etc.
Sophon BM1880 (hybrid ARM + RISC-V)
Sophon BM1880 It is an alternative board developed by Sophon.ia. If you decide to buy one, you will find some features such as:
- 2x Cortex-A53 CPU at 1.5Ghz + RISC-V at 1Ghz
- 1 TPUs @ INT8 thanks to the Tensor processor
- 4GB LPDDR4
- 32GB eMMC flash
- Connectivity Ethernet, WiFi, USB, microSD, Jack, etc.
Intel Neural Stick
Another project similar to the previous ones is this Intel Neural Stick. Version 2 is now available, and the peculiarity in this case is that it is a USB stick that you can easily connect to the PC to start your projects, although it has less versatility than the previous boards. Also, if you need more power, you can use several of them in a USB hub to add capabilities ...
Si purchases this Neural Stick, is priced at about € 100, and is compatible with Linux and Windows. In addition, it allows working with OpenVINO as a development toolkit.
Rockchip you have this powerful hardware accelerated deep learning development kit with which to create very interesting and varied projects. It supports TensorFlow Caffe up to 3 TOPS, as well as Android and GNU / Linux operating systems.
If you want to buy it, you have it available in various versions (ordered from lowest to highest price):
- 3GB of RAM + 16GB of flash
- 3GB of RAM + 16GB of flash + EC25 4G module
- 3GB of RAM + 16GB of flash + 1080p UHD camera
- 6GB of RAM + 32GB of flash + EC25 4G module