Networking & Communications

Embedded Boards Go Deep On Learning, DSP Cores Seek Multi-Gbit Cellular, Wi-Fi Opps

March 16, 2017 By: Mathew Dirjish, Sensors


Embedded Bits, Bytes & Sensors by Mat Dirjish


Embedded Learning

If any platform seems to be all-encompassing, embracing and exploiting every area of technology from simple consumer appliances to complex data analytical systems and medical devices, it's that ever mentioned and constantly looming platform called the Internet of Things (IoT). It's probably safe to say the IoT is here and any developments within can be considered its evolution. Just a few topics we can sandwich under that umbrella of IoT evolution are artificial intelligence, virtual reality, and deep learning. Such is the focus of NVIDIA's latest embedded offering.

The Jetson TX2 is a credit card-sized board that performs computer vision, deep learning, and an array of embedded artificial intelligence tasks (see figure 1). It is offered as an upgrade to the company's Tegra K1 system-on-chip (SoC) based Jetson TK1 and the Tegra X1-based Jetson TX1.

Fig. 1: Jetson TX2 employs NVIDIA's Tegra SoC, based on 64-bit "Denver 2" and ARM Cortex-A57 CPU cores along with a Pascal-generation GPU. TX2 comes in a developer kit (top) and production module (bottom) formats.
Fig. 1: Jetson TX2 employs NVIDIA's Tegra SoC, based on 64-bit "Denver 2" and ARM Cortex-A57 CPU cores along with a Pascal-generation GPU. TX2 comes in a developer kit (top) and production module (bottom) formats.

Production

The production module employs what the company refers to as a "Parker Series SoC. Essentially, its application processor's specs are comparable a SoC.

The Jetson TX2 production module operates in two modes: Max-P (maximum performance) and Max-Q (maximum energy efficiency). In Max-P mode, the board churns out as much as 2X the computational performance of the Jetson TX1 while consuming less than 15W of total module power. Inversely, operating in Max-Q mode, it exhibits up to 2X the energy efficiency of the Jetson TX1 at less than 7.5W of total module power.

Also in Max-P mode, the SoC's GPU runs at 1,122 MHz and the CPU clusters run at 2 GHz when either the Denver 2s or ARM Cortex-A57s are active or 1.4 GHz with both clusters running simultaneously. In Max-Q mode, the GPU runs at 854 MHz and the ARM Cortex-A57 cluster runs at 1.2 GHz with the Denver 2 CPU disabled.

Partners

Developing a variety of IoT applications is a key feature of the Jetson platform. For example, it has enabled Cisco integrate artificial intelligence (AI) features like facial and speech recognition into its Spark products.

Operating in Max-P mode, the Jetson TX2 module successor is able to allegedly double the number of video decodes and encodes and the number of concurrent DNNs and objects capable of being tracked concurrently. It does all this while consuming just 12W of power.

Preferred Modus Operandi

NVIDIA's production module is the company's preferred hardware approach for the majority of users because it enables the company to deliver regular upgrades to its Jetpack SDK based on a known hardware bill-of-materials. Jetpack recently got an upgrade to v3 alongside with the Jetson TX2 release (see figure 2).

Fig. 2: Jetpack SDK bundles software tools for developing computer vision, AI, and other apps for Tegra SoCs (top). Version 3.0 makes advancements in several key SDK elements (bottom).
Fig. 2: Jetpack SDK bundles software tools for developing computer vision, AI, and other apps for Tegra SoCs (top). Version 3.0 makes advancements in several key SDK elements (bottom).

For more details on NVIDIA's Jetson platform, here are a few resources to check out:

Next page

1 2 


About the Author: Mathew Dirjish


Add Comment




Sensors Portal


Automate 2017


IE Europe 2017


IS Auto 2017


MEMS Symposium




Medical Sensors Design Conference




Advertise


Subscribe


Sensorss Expo & Conference 2017 | June 27-29 | San Jose, CA



Twitter Feed

Find It Fix It Forum

Sensors invites you to join the Findit-Fixit Forum, where you can get answers to your sensing questions—concerning technologies, products, methods, applications, and services--and also offer help to your fellow engineers. The Forum covers all kinds of topics, from the basics to the extraordinary.

Join the discussion!


© Copyright 2017 Questex, LLC. All Rights Reserved. Sensorsmag. Privacy Policy | Terms of Use

If you are having technical difficulties or considerations, please contact the webmaster.