Nvidia’s newly announced Jetson TX2 developer kit will offer twice the performance of its predecessor if that’s what hardware partners require. But if they are more interested in energy efficiency, they can have that instead, by using the system in a lower power mode to improve its efficiency by up to two times.
The purpose of Nvidia’s Jetson platform is to offer powerful parallel computing at the local level, to offer “inference at the edge,” as Nvidia puts it. In more layman terms, it means AI processing for things like image classification, navigation, and speech recognition on local hardware, rather than relying on the cloud. That has a myriad of benefits, but most importantly it reduces bandwidth usage, cuts latency, and improves the privacy of data being processed.
Jetson TX2 is looking to do everything its predecessor did and more. It adds a second processor based on Nvidia’s Dual Denver 2 design, alongside the original Quad ARM A57 chip found in the TX1. They are paired with 8GB of LPDDR4, running on a 128-bit interface and offering up to 58.3 GBps of bandwidth — double that of the TX1.
More importantly, the new Jetson system packs in a Nvidia Pascal graphics chip featuring 256 CUDA cores.
The TX2 can encode and decode 4K video at 60Hz and transmit it over an HDMI 2.0 connection. It also comes with 32GB of eMMC onboard storage, plus support for expanded storage from SD and SATA connections.
All of this works out to double the performance potential of the original TX1, but in the same credit card-sized form factor, so hardware partners can simply swap out the TX1 for the TX2 if desired. This lets them offer double the amount of processing that their products can achieve, or lets them operate at a much lower power point. Nvidia claims the TX2 is capable of the same performance of its predecessor while requiring less than 7.5w of power.
All of the hardware improvements of the new Jetson system are backed up by Nvidia’s JetPack 3.0 SDK for AI computing, which lets it support all manner of intelligent systems, with a specific focus on neural networks for image processing and deep-learning capabilities.
Nvidia hopes that this new hardware and software package will help develop new and exciting products for intelligent video analytics like smart cameras, improved functionality in automated vehicles and new, and smarter robots for consumer and business use alike.
The Nvidia Jetson TX2 developer kit can be pre-ordered right now in the U.S. and Europe for $600. Shipping will begin on March 14, while availability in other regions will begin in the coming weeks. The Jetson TX2 module without carrier board will be available in the second quarter of this year with a price tag of $400, though a minimum order of $1,000 is required.
In light of the new generation of hardware, Nvidia has reduced the price of original TX1 developer kit to $500.
Nvidia will show off the TX2 in action at the Embedded World Conference 2017 in Nuremberg, Germany, between March 14 and 16, alongside partners like Cisco, which will showcase a smart whiteboard design making use of the TX2.