Nvidia's Latest AI Module Can Fix Video Conferencing and Lettuce

Artificial intelligence and machine learning are transforming everything from cocky-driving cars to robots that can place how fresh lettuce is, merely their immense processing power requirements take more often than not relegated them to data centers with massive banks of CPUs and GPUs.

So a few companies, including Nvidia, are trying to bring AI processing to the drones, robots, and other devices that can virtually benefit from them. The graphics company on Tuesday unveiled its latest AI compute module, the Jetson TX2, which is designed to eliminate the need to send massive amounts of AI data to and from the cloud.

The Jetson TX2 is an update to Nvidia'due south first GPU for "border" devices, the Jetson TX1, launched in 2022. It has since been adopted by large companies including Toyota and Cisco, equally well as startups like the Blue River lettuce bot company. The TX1 was already powerful, delivering a whopping one teraflop of functioning, but the TX2 delivers twice that processing power. Or, if you lot need power savings, you can become the aforementioned performance as a TX1 with half the ability consumption (7.v watts).

Information technology'south too tiny: the TX2 compute module itself is 2-inches tall and iii.4-inches wide, nigh the height of a double-A battery. Within is a 256-cadre Pascal architecture-based GPU, dual ARM-based CPUs, 8GB of memory, and 32GB of eMMC storage. That compares to the TX1's 4GB of retentivity, 16GB of storage, and a Maxwell-based GPU. Both include HDMI, Ethernet, and back up for decoding up to 2 simultaneous 4K video streams via connected cameras.

With upgraded specs come a $100 price heave. The TX2 programmer kit costs $599, while the TX2 module itself costs $399 in quantities of one,000 or more. The kit allows developers to experiment with their AI algorithms, while the module tin be installed in mass-production products such equally drones or robots. At a press event in San Francisco on Tuesday afternoon, Nvidia Jetson customers showed off some of those devices.

Cisco offered a demo of its Spark Board for 4K video conferencing by subjecting a few of its Norwegian employees to a savage 1:xxx a.one thousand. conference phone call (it was 4:xxx p.m. in California). Using a single 4K video camera, the Jetson-powered Spark Board offered a view of a mostly-empty conference room earlier automatically zooming in on the person who was talking at any given moment. Assuming y'all requite the Spark Board a list of your employees, its Jetson-powered AI can even recognize and display the name of the person talking.

The Spark demo was full of choppy video, nonetheless, drawing attention to a fundamental bandwidth limitation that Jetson can't do anything to solve: even though Jetson does all the 4K video processing locally, the video quality on the other stop is still subject field to bandwidth constraints.

Another Jetson-powered application, the Blue River lettuce-thinning bot, doesn't accept that trouble. Engineers have trained a neural network to scan furrows in a field and place which lettuce plants should exist fertilized and which should exist killed. All of its decision-making ability is centralized in ii onboard Jetson TX1 modules, which control automated fertilizer nozzles. At that place's very little human interaction and no data that has to be sent to the deject for processing.

The company estimates that its bots browse and thin x percent of all the lettuce sold in the US.

Source: https://sea.pcmag.com/components/14412/nvidias-latest-ai-module-can-fix-video-conferencing-and-lettuce

Posted by: bockmartyart49.blogspot.com

Share this post