Nvidia’s Jetson Nano Is an AI Computer for the Masses
A new $99 single-board computer from the GPU giant is aimed at amateur programmers interested in building basic artificial intelligence algorithms.
By Tom Brant
Nvidia’s latest foray into artificial intelligence is a $99 Raspberry Pi-like tiny computer for the masses. Called the Jetson Nano, it’s aimed at tinkerers, hobbyists, and students far removed from the fast-paced, profit-driven world of AI.
Nvidia CEO Jensen Huang declared the company “all-in” when it comes to AI a few years ago, kicking off a succession of multi-thousand-dollar supercomputers that can power through petabytes of information to train machine-learning algorithms. These devices, such as the Tesla P100 and DGX-2, are found in many server farms and data centers, including Facebook data centers. They have their roots in the graphics processing units (GPUs) for which Nvidia is best known.
The new Jetson Nano has a tiny fraction of this power. Its GPU, where most of the AI processing takes place, has 128 Cuda processor cores for a total capacity of 472 gigaflops. This is peanuts compared with the DGX-2, which packs a whopping 16 separate GPUs for a total of 2 petaflops of capacity.
The Jetson Nano is still powerful enough to perform one of the most basic artificial intelligence tasks: processing several streams of video simultaneously. This means it could be used as the brains of a robot tasked with both navigating a factory floor and inspecting the machinery it finds.
It can decode two streams of 4K video, eight streams of 1080p video, or 16 streams of 720p video. It also has a generous port selection, including USB 3.0, DisplayPort, HDMI, and Ethernet. All of this makes it suitable for simultaneously analyzing video feeds from multiple security cameras or sensors aboard a roving self-directed vehicle.
A Raspberry Pi Alternative
Nvidia expects some Jetson Nanos to be used as the brains of such marketable, intelligent robots, but the budget computer’s target audience is tinkerers who are experienced enough to build their own artificial intelligence code. Sometimes referred to as “makers,” they gather at “Maker Faires” worldwide to show off their inventions. Nvidia estimates that there are 30 million makers, meaning a vast potential market for Jetson Nanos.
But it is not an untapped market. Multiple inexpensive, single-board computers already exist, the most well-known of which is the Raspberry Pi. Available in several different versions that cost less than $40, the Raspberry Pi comes with its own version of the Linux operating system, which makes it equally appealing to both programmers and people who just want to build their own ultra-cheap desktop PC.
The Jetson Nano doesn’t come with its own Linux OS, though it does have more I/O options than many Raspberry Pi models. Nvidia is counting on a community of developers to share detailed instructions and source code for their projects.
One early project that programmers worked on while the Jetson Nano was still in development is a simple algorithm for powering small 3D-printed wheeled robots known as “Jetbots.” With an onboard Jetson Nano running a collision-avoidance algorithm from a single AI camera, the tiny bot can perform object detection and collision avoidance simultaneously.
Assuming you’ve got rudimentary coding experience, and perhaps a 3D printer with some spare parts lying around, Nvidia estimates that you could get a similar project up and running in a few hours. Or, you could just use it as a computer without any sensors attached.
If that’s your plan, you might be interested in upgrading to the $129 version of the Jetson Nano, which will start shipping by the end of June and comes with 4GB of flash memory included. The $99 version available now doesn’t have any form of storage or memory, so you’ll need to bring your own SD card or USB stick.
Originally published at www.pcmag.com.
