Nvidia’s $399 Jetson Xavier NX Developer Kit runs Linux on a six-core Jetson Xavier NX processor with AI performance up to 21 TOPS. The company also announced integrated support for NX and other Jetsons in the cloud, enabling the creation of container-driven applications.
When Nvidia introduced its Jetson Xavier NX computer module last November, there were no finished developers, as was the case with the lower Jetson Nano module. This allowed several third-party NX media to find a niche. Nvidia has just released a $399 Jetson Xavier NX development kit that combines the Jetson Xavier NX with a feature set and layout for $99.
Jetson Xavier NX Development Kit (left) and sitting next to the Jetson Xavier NX
module (click on images to enlarge)
The price difference reflects the high CPU and GPU performance of the Jetson Xavier NX and perhaps the pre-integrated Linux software stack, which for the first time includes integrated cloud support for containerized applications. During Nvidia’s preliminary discussion on hiring developers a few weeks ago, the hosts barely mentioned the motherboard, but concentrated on the news that Nvidia has now added integrated cloud support for all of its Jetson modules.
This means that the Jetson TX2, which roughly replaces the NX with a 10-fold increase in performance, can also handle containers. The same applies to the high-end JetSon AGX Xavier, whose NX uses Volta GPU technology (but with 384 GPU cores instead of 512). Native support for cloud computing will also be available on the Jetson Nano, from which the NX derives its compact 69.6 x 45 mm base.
Jetson Xavier NX
(click on image to enlarge)
The Xavier NX-Hex kernel supports up to 21 TOPS at 15 W or 14 TOPS at 10 W of neural processing power. By comparison: AGX Xavier achieved 32 TOPS.
According to AnandTech’s table released today with the Developer Bundle Report, 6x Carmel Armv8.2 processor cores at 1.4 GHz can work with 15 W or only 2x cores at 1.9 GHz. At 15 W, the Volta’s GPU operates at 1.1 GHz. At 10 W, the NX is limited to a 4x 1.2 GHz active CPU core or a 2x 1.5 GHz core with a Volta GPU clock speed up to 800 MHz.
In short: Here is the current composition of Nvidia Jetson:
- Jetson Nano – 69.6 x 45 mm; 4x-A57 @ 1.43 GHz CPU; 128-core Maxwell GPU; 4 GB LPDDR4; 16 GB eMMC
- Jetson TX2 – 87 x 55 mm; 2x Denver and 4x -A57 CPU; 256-core Pascal GPU; 8 GB LPDDR4; 32 GB eMMC; selection. WiFi/BT
- Jetson Xavier NX – 69.6 x 45 mm; 6x ARMv8.2 Carmel-CPU; 384-core Volta-GPU; 48 tensor cores; 8GB LPDDR4; selection. 8 GB to 32 GB eMMC
- Jetson Xavier AGX – 105 x 87 mm; 8x ARMv8.2 Carmel (-A75 similar); 512 core Volta GPU; 64 sensor cores; 16 GB LPDDR4; 32 GB selectable EMC
Nvidia’s new cloud support will help manufacturers and developers implement common improvements, improve accuracy and take advantage of the latest features of Jetson-based AI devices, Nvidia said. And developers can quickly implement new algorithms on a large scale throughout the application life cycle, while minimizing downtime.
Cloud support also means that developers can use containerized artificial intelligence applications used on higher grade Nvidia GPUs for the data center. Nvidia has developed an incredible amount of artificial intelligence software that has been optimized for different industries, and now, with support from the cloud, customers can take advantage of this software innovation through our powerful and energy efficient Jetson family, says Deepu Talla, Vice President and General Manager of Computer Peripherals at Nvidia.
The NX Development Kit comes with the Ubuntu-based Linux stack, which contains the Nvidia CUDA-X accelerated computing stack and the Jetpack SDK, which comes with the Nvidia container for executables. The kit also provides access to the Nvidia NGC register, which contains various containerised and prepared AI models with docker and singularity functions.
Developers can work with NGC models and the latest Nvidia tools and optimize them directly on the Jetson Xavier NX and other Jetson development kits, according to Nvidia. You can also use a workstation with cross tool chain compilation or even cloud-based development workflows with containers that will soon arrive at the CNG.
Jetson Xavier NX function (left) and diagrams compare AI markers with other Jetson
modules (click on the images to enlarge).
NGC Container applications support in-depth HPC training and visualization based on TensorFlow, PyTorch, MXNet, Nvidia TensorRT, RAPIDS and other frameworks. In the foreword, Nvidia emphasized the support of Jetson Xavier NX for Google’s BERT (bi-directional representation of the transformative encoder), a technique for natural language processing (NLP) in neural networks, which Google uses in its search engine.
Nvidia also showed a robot driven by NX that simultaneously analyzes the input of 4 cameras, each controlled by separate containers. The whole computer runs on 15 W. The robot can count people and the distance between them and can also use algorithms to recognize a pose and a view. He can also interact with people by analyzing natural language attitudes.
Jetson Xavier NXHardware Development Kit
Although the specifications of the Jetson Xavier NX Development Kit are less detailed than those of the Jetson Nano Development Kit, the NX appears to be virtually identical in terms of equipment and layout. The card is slightly larger than the Nano dev 103 x 90.5 x 31 mm kit versus 100 x 80 x 29 mm, and the Jetson module is cooled by a fan instead of a cooler.
Jetson Xavier NX Developer Kit (left) and Jetson Nano
Developer Kit (click on the images to enlarge).
In addition to the more powerful Jetson module, an M.2M key lock with NVMe support is an important addition to the new devices. It is connected to an old M.2 E key slot, which here is equipped with a WiFi/Bluetooth module.
The NX kit contains 2x MIPI-CSI instead of MIPI-CSI tape and DisplayPort instead of eDP. Instead of supporting one 4K display in the Nano SDK, you can run two 4Kp60 displays simultaneously or up to 32x 1080p30 streams. The video coding now works from 2x 4Kp30 to 14x 1080p30.
The 4 USB 3.1 ports are probably Gen1 ports, as in the Nano Development Kit. There were no details about the DC connection of the board of directors. Tom’s hardware report indicates that the micro-USB port can no longer power the card that came with the 65W camera, but is likely to be available to end users with a 45W camera.
Again, the card has a 40-pin GPIO connector that works with Raspberry Pi HAT if you use the right Python library, says Tom’s hardware. The story also says that the MIPI-CSI connector supports Raspberry Pi cameras, including a new high quality 12.3 megapixel camera.
The 16 GB of eMMC equipment found on the NX production module is not available here, as Tom’s hardware confirms. The production module is sold individually for $459 – $60 more than the included version for $399. The price difference is due to the superior quality of the parts of the production module, with a lifetime warranty of 5 to 10 years, depending on the history.
Limited specifications for the Jetson Xavier NX Development Kit included:
- Processor (via Jetson Xavier NX) – 6x ARMv8.2 Carmel CPU; 6 MB L2, 4 MB L3 cache; 384-core graphics processor Volta with 48 sensor cores; 2x NVDLA engines; 7-way VLIW vision processor; artificial intelligence support CUDA-X.
- Memory/storage :
- 8 GB 128-bit LPDDR4 with 51.2 GB/s (via Jetson Xavier NX)
- MicroSD lock
- The NVMe drive is available on the M.2 M flash drive (see extension below).
- Gigabit Ethernet connection
- WiFi/BT installed on E-Key M.2 (see extension below)
- Media input/output :
- HDMI connection
- H.265 video decoding – 2x 4Kp60, 4x 4Kp30, 12x 1080p60, 32x 1080p30 (H.265)
- H.264 video decoding – 2x 4Kp30, 6x 1080p60, 16x 1080p30
- 2x MIPI CSI-2 D-PHY tracks
- H.265/H.264 video coding – 2x 4Kp30, 6x 1080p 60, 14x 1080p30
- I2S audio
- Other inputs/outputs :
- 4x USB 3.1 host ports
- Connection for Micro-USB 2.0 device
- 40-pin header with GPIO, I2C, SPI, UART.
- M.2 M-Key lock with NVMe support
- M.2 Lock for key E with WiFi/BT pre-programmed module
- Other topics – Fan
- Dimensions – 103 x 90.5 x 31 mm
- Operating system – Linux4Tegra (Linux 4.9) based on Ubuntu, with Jetpack 4.2 SDK.
The Jetson Xavier NX Development Kit is available for $399, including the Xavier NX module. You can find more information on Nvidia’s announcements and products/activities page.