Data Science

NVIDIA Jetson TX2: The New Gold Standard for AI at the Edge

The newest member of the Jetson family — Jetson TX2 — offers a comprehensive solution to challenges faced by developers looking to push the boundaries of AI at the edge.

The credit card-sized Jetson TX2 is the world’s leading high-performance, low-power embedded platform. It features a 256-core NVIDIA Pascal GPU, a hex-core ARMv8 64-bit CPU complex, and 8GB of LPDDR4 with a 128-bit interface. TX2’s CPU complex includes a dual-core 7-way superscalar NVIDIA Denver 2 for high single-thread performance with dynamic code optimization, and a quad-core ARM Cortex-A57 geared for multithreading.
Jetson TX2 offers two modes of operation that allow developers to tailor performance and power consumption to a diversity of application demands. In Max-Q preset mode, Jetson TX2 operates at peak energy efficiency in an envelope of less than 7.5W, using half the power for the same performance as Jetson TX1. Max-P, the other preset platform configuration, enables maximum system performance in less than 15W. This means that the Jetson TX2 is twice as powerful as the Jetson TX1 (here’s a more detailed analysis).
Jetson TX2 is supported by the JetPack 3.0 SDK. Built on Linux kernel 4.4 and Ubuntu 16.04 LTS, the JetPack SDK makes it easier to develop AI applications with tools such as CUDA 8, cuDNN, TensorRT, VisionWorks, Gstreamer, and OpenCV. Developers can train deep learning models in the cloud, datacenter, or PC with GPU-accelerated NVIDIA DIGITS 5 and deep learning frameworks like Caffe, Torch, Theano, and TensorFlow. Trained models can then be deployed to Jetson TX2 with the TensorRT inference engine to maximize throughput and efficiency. Jetson also benefits from NVIDIA’s consistent software and hardware platform across embedded, PC and server environments for a seamless development experience from application development to edge deployment.

To get started, NVIDIA provides the Jetson TX2 Developer Kit, which includes a reference carrier board and a 5 megapixel CSI-2 camera module. The developer kit provides PC connectors including USB 3.0, HDMI, RJ45 Ethernet, SD card, and a PCIe x4 slot. Developers also have access to open documentation, design schematics, software updates, and forums support through the NVIDIA developer website. While the documentation and design references allow anyone to create Jetson compatible cameras, carrier boards, and other accessories, the Jetson platform’s strong ecosystem partners also provide off-the-shelf solutions.
To test out the developer kit, NVIDIA’s Two Days to a Demo tutorial enables developers to start deploying deep learning on Jetson. This guide supports Jetson TX2 and TX1, and now includes new features like segmentation networks. The code is available on GitHub along with step-by-step instructions to illustrate the ease of deploying AI on Jetson.
You can pre-order the NVIDIA Jetson TX2 Developer Kit today for $599 in the United States. Shipping will begin March 14, and Jetson TX2 will be available in other regions in the coming weeks.  The Jetson TX2 module will be available in Q2 for $399 (in quantities of 1,000 or more) from NVIDIA and its distributors around the world. The price of the Jetson TX1 Developer Kit is now $499.
Start developing with Jetson TX2. Pre-order here.
Learn all about Jetson TX2 on the Parallel Forall blog.

Discuss (0)