[Updated: May 18] — Nvidia’s $399 Jetson Xavier NX Developer Kit runs Linux on the hexa-core Jetson Xavier NX with up to 21 TOPS AI performance. The company also announced cloud native support for the NX and other Jetsons, thereby enabling container apps.

Last November, when Nvidia unveiled its Jetson Xavier NX compute module, there was no maker-friendly developer kit at the ready as there had been with the lower-end Jetson Nano. This left room for a variety of third-party NX carriers to find a niche. Now, Nvidia has launched a $399 Jetson Xavier NX Developer Kit that combines the Jetson Xavier NX with a feature set and layout that is similar to the $99 Jetson Nano Dev Kit. Nvidia also announced a new high-end Ampere GPU technology and A100 GPU that includes an edge AI version called the EGX A100 (see farther below).

Jetson Xavier NX Developer Kit (left) and sitting next to Jetson Xavier NX module
(click images to enlarge)

The price difference represents the greater CPU and GPU power of the Jetson Xavier NX, and perhaps also the pre-integrated Linux software stack that for the first time includes cloud native support to enable container applications. In an Nvidia prebrief on the dev kit a few weeks ago, the presenters barely mentioned the carrier board, but focused on the news that Nvidia has now added cloud native support for all its Jetson modules.

That means the Jetson TX2, which the NX is essentially replacing with a claimed 10x performance leap, can also run containers. The same goes for the high-end Jetson AGX Xavier, from which the NX borrows its Volta GPU technology (albeit with 384 GPU cores instead of 512). Cloud native support will also be available on the Jetson Nano, from which the NX borrows its compact 69.6 x 45mm footprint.

Jetson Xavier NX
(click image to enlarge)

The hexa-core Xavier NX supports up to 21 TOPS neural processing performance at 15W, or 14 TOPS at 10W. This is compared to 32 TOPS for the AGX Xavier.

According to an AnandTech chart posted today with its dev kit report, the 6x Armv8.2 Carmel CPU cores can run at 1.4GHz at 15W or only 2x cores can run at 1.9GHz. At 15W, the Volta GPU runs at 1.1GHz. At 10W, the NX is limited to 4x active 1.2GHz CPU cores or 2x 1.5GHz cores, with the Volta GPU clocked to 800MHz.

To recap: here is the current Nvidia Jetson line-up:

• Jetson Nano — 69.6 x 45mm; 4x -A57 @ 1.43GHz CPU; 128-core Maxwell GPU; 4GB LPDDR4; 16GB eMMC
• Jetson TX2 — 87 x 55mm; 2x Denver and 4x -A57 CPU; 256-core Pascal GPU; 8GB LPDDR4; 32GB eMMC; opt. WiFi/BT
• Jetson Xavier NX — 69.6 x 45mm; 6x ARMv8.2 Carmel CPU; 384-core Volta GPU; 48 tensor cores; 8GB LPDDR4; opt. 8GB to 32GB eMMC
• Jetson Xavier AGX — 105 x 87mm; 8x ARMv8.2 Carmel (-A75 like); 512-core Volta GPU; 64 tensor cores; 16GB LPDDR4; opt. 32GB eMMC

Nvidia’s new cloud-native support “helps manufacturers and developers implement frequent improvements, improve accuracy and use the latest features with Jetson-based AI edge devices,” says Nvidia. “And developers can quickly deploy new algorithms throughout an application’s lifecycle, at scale, while minimizing downtime.”

Cloud-native support also means developers can leverage containerized AI applications that are being used on higher end Nvidia GPUs for the datacenter. “Nvidia has created an incredible amount of optimized AI software across multiple industries, and now, with cloud-native support, customers can leverage that software innovation through our high-performance, low-power Jetson family,” stated Deepu Talla, VP and GM of edge computing at Nvidia.

The NX dev kit ships with an Ubuntu-based Linux stack that includes the Nvidia CUDA-X accelerated computing stack and the Jetpack SDK, which comes with an Nvidia container runtime. The kit also provides access to Nvidia’s NGC Registry, which hosts several containerized, pre-trained AI models with Docker and Singularity runtimes.

“Developers can work with the models from NGC and the latest Nvidia tools and optimize directly on a Jetson Xavier NX developer kit or other Jetson developer kits,” says Nvidia. “They can also use a workstation with cross-compile toolchain or even cloud development workflows using containers coming soon to NGC.”

Jetson Xavier NX feature (left) and AI benchmark comparison charts with other Jetson modules
(click images to enlarge)

The NGC container applications support deep learning and HPC visualization based on TensorFlow, PyTorch, MXNet, Nvidia TensorRT, RAPIDS and other frameworks. In the prebrief, Nvidia emphasized Jetson Xavier NX support for Google’s BERT (Bidirectional Encoder Representations from Transformers), a neural network-based technique for natural language processing (NLP) pre-training that Google has applied to its search engine.

Nvidia also demonstrated an NX-powered robot that analyzes inputs from 4x cameras simultaneously, each of which is controlled via separate containers. The entire computer runs on 15W. The robot can count people and the distance between them and apply pose and gaze detection algorithms. It can also interact with people via natural language pose analysis.

Nvidia EGX A100 deployed via Mellanox ConnectX-6 Dx SmartNIC
(click image to enlarge)

(Update: Following the NX dev kit announcement, Nvidia unveiled a new Nvidia Ampere GPU technology and an Nvidia A100 GPU based on it that can be partitioned into 7x different GPUs and is claimed to boost performance by up to 20x over its predecessors. Although aimed primarily at the datacenter, a relatively low end EGX A100 version is targeted at edge AI processing. EGX A100 is being initially deployed via an Nvidia Mellanox ConnectX-6 Dx SmartNIC board.)

 
Jetson Xavier NX Developer Kit hardware

Although the specs listed for the Jetson Xavier NX Dev Kit are less detailed than those posted for the Jetson Nano Dev Kit, the NX board appears to be almost identical in features and layout. The board is slightly larger than the Nano dev kit at 103 x 90.5 x 31mm compared to 100 x 80 x 29mm, and the Jetson module is cooled by a fan instead of a heatsink.

Jetson Xavier NX Dev Kit (left) and Jetson Nano Dev Kit
(click images to enlarge)

Aside from the more powerful Jetson module, the “key” addition to the new dev kit is an M.2 M-key slot with NVMe support. This joins the earlier M.2 E-key slot, which is here populated by a WiFi/Bluetooth module.

The NX kit has 2x MIPI-CSI lanes instead of one and a DisplayPort instead of eDP. Instead of supporting a single 4K display on the Nano dev kit, you can run two 4Kp60 displays simultaneously, or up to 32x 1080p30 streams. Video encode, meanwhile, runs from 2x 4Kp30 to 14x 1080p30.

The 4x USB 3.1 ports are are likely Gen1, as on the Nano dev kit. There were no details about the board’s DC jack. A Tom’s Hardware report says that the micro-USB port can no longer power the board, which shipped to reviewers with a 65W brick, but will likely be available to end users with a 45W brick.

Once again, the board has a 40-pin GPIO connector, which will work with Raspberry Pi HATs “if you use the proper Python library,” says Tom’s Hardware. The story also says the MIPI-CSI connector supports Raspberry Pi cameras including the new, 12.3-megapixel High Quality Camera.

The 16GB eMMC found on the production NX module is not available here, confirms Tom’s Hardware. The production module sells for $459 in individual samples — $60 more than the version included with the $399 dev kit. The price difference is due to the production module’s higher-quality components with 5-10 year lifespan guarantees, says the story.

Limited specifications for the Jetson Xavier NX Developer Kit include:

• Processor (via Jetson Xavier NX) — 6x ARMv8.2 Carmel CPU; 6MB L2, 4MB L3 cache; 384-core Volta GPU with 48 tensor cores; 2x NVDLA Engines; 7-Way VLIW Vision Processor; CUDA-X AI support
• Memory/storage:
  • 8GB 128-bit LPDDR4 @ 51.2 GB/s (via Jetson Xavier NX)
  • MicroSD slot
  • NVMe storage available on M.2 M-key (see expansion below)
• Networking:
  • Gigabit Ethernet port
  • WiFi/BT installed on M.2 E-key (see expansion below)
• Media I/O:
  • HDMI port
  • DisplayPort
  • H.265 video decode — 2x 4Kp60, 4x 4Kp30, 12x 1080p60, 32x 1080p30 (H.265)
  • H.264 video decode — 2x 4Kp30, 6x 1080p60, 16x 1080p30
  • 2x MIPI CSI-2 D-PHY lanes
  • H.265/H.264 video encode — 2x 4Kp30, 6x 1080p 60, 14x 1080p30
  • I2S audio
• Other I/O:
  • 4x USB 3.1 host ports
  • Micro-USB 2.0 device port
• Expansion:
  • 40-pin header with GPIOs, I2C, SPI, UART
  • M.2 M-key slot with NVMe support
  • M.2 Key E slot with pre-installed WiFi/BT module
• Other features — fan
• Dimensions — 103 x 90.5 x 31mm
• Operating system — Linux4Tegra (Linux 4.9) based on Ubuntu, with Jetpack 4.2 SDK

 
Further information

The Jetson Xavier NX Developer Kit is available for $399 including the Xavier NX module. More information may be found in Nvidia’s announcement and product/shopping page.