nvidia's jetson tx2 takes machine learning to the edge - white board sizes

Small development boards have become a key enabler for the latest wave of hardware startups.
The most popular boards such as Arduino and Raspberry Pi inspire many projects and many knocksoffs.
These boards can start a project like a beer brewing machine, but they have limited computing power.
Where do you go when your project needs real computing power and maybe some local machine learning?
The NVIDIA Jetson board is a popular choice.
The first board is the TX1 released on November 2015.
Now NVIDIA has released more powerful features-
Efficient Jetson TX2 board.
Jetson board is the brother of NVIDIA's Drive PX board for autonomous driving, and TX2 shares the same Tegra "Parker" silicon as Drive PX 2.
There are many synergies between the two families, as both can be used to add local machine learning to transport.
The Drive PX board is designed for automotive with extended temperature range and high reliability requirements.
The Jetson board is optimized for compact housing and battery power for small portable devices.
Equipped with robots, drones, 360 cameras, medical and other equipment.
Jetson can be used for "edge" machine learning.
When the connection bandwidth is limited or unstable, it is useful to be able to process data at local and limited power (
Like in remote areas)
Delay is crucial (real-time control)
Or privacy and security is a problem.
At the Jetson TX 2 conference, NVIDIA held an event in San Francisco, attracting some companies and researchers who use Jetson in their products or projects.
The star of the show is the new Cisco Spark Board, which combines a large touch-sensitive display as an interactive whiteboard that can serve as a video conferencing system for identifying and adapting people in the room.
The board is able to track the number of people in the room and adjust the display zoom and offset accordingly.
It's also 12-
Element microphone array to track who is talking and highlight the person.
In addition, the board can also wake up automatically when authorized personnel enter the room and resume work from the last location.
Combining whiteboards, collaboration tools, and video conferencing on one screen can improve the efficiency and productivity of the system.
The integrated Jetson supports the 4 k resolution display of the Spark Board and can help with local face recognition.
One of my favorite products displayed using Jetson board is the portable handheld 3D scanner for Artec 3D.
The "Leo" handheld scanner can be used to create a 3D model of any object until the size of the car.
An old version was used to make a 3D model of President Obama.
The Jetson board is used to calculate and display parts in order to successfully model objects on the display on the back of the scanner.
The image forms a point-
The cloud of the object, which later becomes a set of geometry and textures for building 3D models in a computer.
The level of detail is set by the user.
The final result is a detailed 3D model that can be used in art and design, including CGI, healthcare, industrial design, science and education, etc.
Another interesting system to use Jetson is the intelligent drone on the way to automatic search and rescue operations and maintenance and inspection work.
Drones use deep learning on Jetson for target detection, path planning, and collision avoidance.
The MIT representative is a Jetson.
Mobile research robots.
Toyota also has a research robot using Jetson's "Human Support.
Compared to the original TX1, the main advantage of Jetson TX2 is that it adds two extra, higher performance Denver CPU cores to four Cortex
The A57 kernel in NVIDIA's latest Pascal GPU TX1 provides twice the memory capacity and bandwidth. (
For more details, see the table below. )
Because Jetson TX2 is based on the same Tegra chip used in the NVIDIA Drive PX2 platform, it uses the car-grade 16nm FinFET process for TSMC.
Newer process nodes help reduce the power of the 20nm process used on the tx1.
The rated power of the TX1 is 10 w, but the TX2 can provide similar performance in less than 7. 5W (
What NVIDIA calls Max-Q).
TX2 can also run at full speed at 15 w (Max-P)
The calculation performance of Tx1 is twice that of tx1.
The CPU clock speed is up to 2 GHz and The Pascal GPU can run up to 1. 3GHz.
The company provides JetPack software development kits for Developers (SDK)
Use the company's extensive AI ecosystem for machine learning and GPU computing.
The SDK supports Google's Tensorflow, as well as NVIDIA's cuDNN and CUDA libraries.
The operating system of the Development Board is Ubuntu Linux.
The latest JetPack 3 SDK includes an Ubuntu Developer Preview with kernel 4. 4.
The Jetson TX2 Developer Suite, which began shipping in March 14, is available from Internet retailers such as NVIDIA and NewEgg.
The TX2 board is not cheap for $599, but the production board is $399 (
Quantity is 1000).
NVIDIA will not eliminate the TX1 motherboard, but will reduce the price of the Jetson TX1 development kit to $499.
In addition, NVIDIA reduced the price of Jetson TX1 production board to $299, the company promised 5-year lifespan.
Since the TX2 board is a direct replacement for the TX1, developers who have started using the TX1 can quickly migrate to the TX2 to save power or get more performance.
For projects with limited power and space, TX2 provides the best GPU-
The auxiliary machine learning platform on the market today.
For these applications, although the power and size of the board are small, the software ecosystem is almost more important.
Kevin KrewellPrincipal, chief analyst at TIRIAS Research--
The authors and members of the TIRIAS researchers do not hold equity positions in any of the above companies.
TIRIAS Research is tracking and consulting companies from semiconductors to systems, sensors to the entire electronic ecosystem of the cloud.