NVIDIA Jetson Xavier Announcement

There is a lot of news in the Jetson world today. I know that a lot of it may be obscured by the big news that the JetsonHacks YouTube Channel just passed over one million views. Thanks for watching! We will cover the other events here just so they are not drowned out.

Jetson Xavier

Note: July, 2018 – Phil Lawrence from NVIDIA gives a walk through of the Jetson Xavier! Looky here:

Note: June 30, 2018 – Technical Specifications Published!

At Computex 2018, NVIDIA announced the Jetson Xavier, the latest addition to the Jetson platform family. Money quote:

Jetson Xavier is designed for robots, drones and other autonomous machines that need maximum compute at the edge to run modern AI workloads and solve problems in manufacturing, logistics, retail, service, agriculture and more.

Here’s a beauty shot:

Jetson Xavier
Jetson Xavier (courtesy NVIDIA)

Specs

Here’s some of the technical specs:

System

  • GPU — 512-core Volta GPU with Tensor Cores
  • DL/ML Accelerator — (2x) NVDLA Engines (nvdla.org)
  • CPU — 8-core ARMv8.2 64-bit CPU, 8MB L2 + 4MB L3
  • Memory — 16GB 256-bit LPDDR4x | 137 GB/s
  • GPU L2 cache size is 512KB. There’s also a 4MB L3 cache shared between CPU/GPU
  • Storage — 32GB eMMC 5.1
  • Vision Accelerator — 7-way VLIW processor
  • Video Encode — (2x) 4Kp60 | HEVC
  • Video Decode — (2x) 4Kp60 | 12-bit support
  • Mechanical — 100mm x 87mm with 16mm Z-height
    (699-pin board-to-board connector)
  • Multiple operating modes at 10W, 15W, and 30W
  • weight of the devkit is 630 grams

I/O

  • Display — 3x eDP/DP/HDMI at 4Kp60 | HDMI 2.0, DP HBR3
  • Camera — 16x CSI-2 Lanes (40 Gbps in D-PHY V1.2 or 109 GBps in CPHY v1.1
    • 8x SLVS-EC lanes (up to 18.4 Gbps)
    • Up to 16 simultaneous cameras
  • PCIe — 5x 16GT/s gen4 controllers | 1×8, 1×4, 1×2, 2×1
    • (3x) Root Port + Endpoint
    • (2x) Root Port
  • USB
    • (3x) USB 3.1 (10GT/s)
    • (4x) USB 2.0 Ports
  • Ethernet — Gigabit Ethernet-AVB over RGMII
  • Other I/Os — UFS, I2S, I2C, SPI, CAN, GPIO, UART, SD

Jetson Xavier Developer Kit:

Jetson Xavier Module
Jetson Xavier Development Kit (courtesy NVIDIA)

Discussion

Looking over the specs, we can see a couple of interesting items. First, there are 512 Volta based CUDA cores and 8 ARM CPU cores. NVIDIA is quoting around 10 TFLOPS of FP32 performance! That equates to up to 20x the performance of the original Jetson TX2! Now, the Xavier can use up to 30W (which is 2X the Jetson TX2) in max mode which accounts for some of the speed increase. The Jetson Xavier has 15W and 10W modes too.

A 256 bit wide memory bus also helps (the Jetson TX2 is 128). The 16GB memory is twice the Jetson TX2, and the eMMC is the same size as the TX2. I/O grows quite a bit, now there are 3 USB 3.1 ports on the module, along with 4 2.0 ports. You can use up to 16 simultaneous cameras.

PCIe is Gen 4! This is faster than any currently shipping desktop systems.

The Jetson Xavier Module is 100mm x 87mm with 16mm height. The Jetson Xavier DevKit Reference Carrier is only slightly larger and adds a large variety of I/O connectors such as USB and HDMI. The DevKit has a CSI camera slot, along with a PCIe connector and access to the GPIO connector. This is quite a change from previous generations with their much larger carrier board.

The Jetson Xavier does not have onboard Wi-Fi or Bluetooth, a change from the earlier Jetsons TX modules. However the Xavier Devkit has a M.2 Key E slot which will allow developers to easily add WiFi/BlueTooth/Cell connectivity.

There is a M.2 Key M slot for SSD storage located on the top of the carrier board (right beneath the module). The SD card is micro-SD format. There is also an eSATAp port.

In addition to powering the Jetson Xavier DevKit via the barrel jack (which is the same size as the Jetson TX1/TX2), the device can be powered via USB C.

Pricing at the announcement the MSRP of the Jetson Xavier Developer Kit is $1299 (USD). The Jetson Xavier Developer Kit will be available for early access in August and starting in October for general access.

Isaac

Also announced at Computex 2018 is NVIDIA® Isaac™ availability. Money quote:

The NVIDIA Isaac Software Development Kit (SDK) gives you a comprehensive set of frameworks, tools, APIs, and libraries to accelerate development of robotics algorithms and software.

Isaac Robotics Software

NVIDIA provides a toolbox for the simulation, training, verification and deployment of Jetson Xavier. This robotics software consists of:

  • Isaac SDK – a collection of APIs and tools to develop robotics algorithm software and runtime framework with fully accelerated libraries.
  • Isaac IMX – Isaac Intelligent Machine Acceleration applications, a collection of NVIDIA-developed robotics algorithm software.
  • Isaac Sim – a highly realistic virtual simulation environment for developers to train autonomous machines and perform hardware-in-the-loop testing with Jetson Xavier.

Conclusion

New stuff in the pipe! Mo’ better, mo faster. It promises to be a fun fall when we get out hands on these little beasties.

Updates

Updates since the original article:

  • June 30, 2018 – Technical Specifications Published!
  • Confirmation of Gen 4 PCIe
  • Clarification of Jetson Xavier Development Kit (It’s pictured above)
  • Clarification that Xavier has twice the amount of memory as the TX2
  • Development Kit has a M.2 Key E slot for wireless cards
  • Carrier board has a M.2 key M for SSD storage on the top of the carrier
  • SD Card is micro format
  • eSATAp port
  • Development Kit has a CSI slot, similar to the Jetson TX1/TX2. There is no camera included in the Dev Kit.
  • Jetson Xavier can be powered over USB C as well as through the barrel jack. The barrel jack is the same size as the Jetson TX1/TX2.

4 Comments

    • Here’s what they’re quoting:

      Jetson Xavier is capable of more than 30 TOPS (trillion operations per second) for deep learning and computer vision tasks. The 512-core Volta GPU with support for Tensor Cores and mixed-precision compute is capable of up to 10 TFLOPS FP16 and 20 TOPS INT8. Jetson Xavier’s dual NVDLA engines are capable of up to 5 TOPS each. It also has high-performance eight-core ARM64 CPU, a dedicated image processor, a video processor and a vision processor for accelerating computer vision tasks.

      Thanks for reading!

Leave a Reply

Your email address will not be published.


*