NVIDIA announces Jetson Xavier NX

We are excited to introduce Jetson Xavier NX, the newest member of the Jetson family!

Jetson Xavier NX is based on the Xavier SoC and is pin-compatible with the Nano form-factor. It has 10/15W power modes, with up to 21 TOPS of AI performance. High-level features and specifications include:

  • 384-core Volta GPU with 48 Tensor Cores
  • 6-core NVIDIA Carmel 64-bit ARMv8.2 CPU
  • 8GB 128-bit LPDDR4x @ 1600MHz | 51.2GB/s
  • Dual NVIDIA Deep Learning Accelerators
  • 260-pin SODIMM edge connector, 70x45mm
  • 10W / 15W modes, 5V input

The Jetson Xavier NX module is available for order now for $399 USD in volume and will be available March 2020 the 2nd half of April 2020. You can also get started today with the Jetson AGX Xavier Developer Kit and a patch for JetPack that emulates the configuration of the Jetson Xavier NX. Documentation and design collateral for Jetson Xavier NX is also available to download through the URL below and the following list:

Module Documentation

Software Downloads and Documentation

For more info and benchmarks, check out our article Introducing Jetson Xavier NX, the World’s Smallest AI Supercomputer

Page…https://developer.nvidia.com/embedded/jetson-xavier-nx
Docs…https://developer.nvidia.com/embedded/downloads#?tx=$product,jetson_xavier_nx
Blog…https://devblogs.nvidia.com/jetson-xavier-nx-the-worlds-smallest-ai-supercomputer

Great!

This appears to be similar to an Xavier without the Denver cores. Is this a good description?

TX2 was the one with the quad-core A57 and two Denver cores. Xavier uses all Carmel CPU cores (the next version of Denver)

Great news, will there be a thermal solution integrated or a “devkit” model will be shipped?

As with the commercial Nano module, the Xavier NX module will ship without a TTP, and like previous Jetson’s we will make a thermal design guide available in addition to the module CAD model which is already available. Ecosystem partners may also offer cooling solutions for the Xavier NX module like they have with the Nano module.

We don’t have info to share about a devkit at this time, but for now it is recommended to get started with the Jetson AGX Xavier devkit for software development using the emulation patch, and since Xavier NX is pin-compatible with Nano, you can use the Nano form-factor as a hardware reference in addition to the Xavier NX hardware design docs and collateral listed above.

Thank you for these details. always good to know ;)

3D Model of the new Jetson Xavier NX available here :

https://skfb.ly/6OH8y

https://sketchfab.com/Fa_Sketch/collections/nvidia

Would anyone explain the reason why the Xaiver NX is better AI DL performance than Xavier 8GB even though the number of CPU and GPU cores are the same and the memory performance of Xavier is better than NX?
NX: 21 TOPS
Xavier 8GB: 20 TOPS
https://developer.nvidia.com/embedded/develop/hardware

Best regards.
Kaka

Will I be able to use the Jetson Xavier NX board with the Jetson Nano Developer Kit Carrier Board?

Hi Dustin!

Does Xavier NX have a SPE Cortex R5 for realtime work?

Looks like a Xavier AGX 8GB but with better power efficiency, only 16GB eMMC and the 260-pin SO-DIMM connector from the Jetson Nano, right?

Hi Kaka, the GPU on Jetson Xavier NX has a higher operating frequency of 1100MHz vs 905MHz for Jetson AGX Xavier 8GB.

Jetson Xavier 8GB has 256-bit memory and 20W power, whereas Xavier NX has 128-bit memory and 10/15W power but higher GPU clocks than Xavier 8GB.

dusty_nv

Thank you for your information! I understand.

Best regards.
Kaka

Thanks for the information. She helped me.

The Jetson Xavier NX can encode video at 2x4K@30. Can it be configured to encode a single 4K@60?

Best Regards,
Ken

What is the weight for the NX module

Hi efiryw2d, not sure yet what the final weight of the production NX module will be.

Hi Ken, Jetson Xavier NX has two video encoder engines, each capable of encoding up to 464MP/s of H.265 (HEVC)

Since 4Kp60 is ~497MP/s, this would be over the limit that an encoder could handle. For more info, please see section 1.7.2 of the datasheet.