NVIDIA is pleased to announce Jetson AGX Xavier, the latest addition to the Jetson platform with greater than 10x the energy efficiency and more than 20x the performance of it’s predecessor, Jetson TX2. It’s an AI computer for autonomous machines, delivering the performance of a GPU workstation in an embedded module under 30W. With multiple operating modes at 10W, 15W, and 30W, Jetson Xavier has a peak performance of more than an incredible 30 TOPS (teraops) of mixed-precision FP32/FP16/INT8 performance. Please refer to the FAQ for further information.
US, Canada, UK, France, Germany, Italy, Poland, Spain, and China
Additionally, orders can be placed through distributors for:
Japan, Europe, Israel, and Turkey (other regions to follow soon)
Initial orders will be fulfilled on a first-come first-served basis within an estimated six-week time period beginning early September 2018.
The price of the devkit for Registered Developers is $1,299 (limit 1 unit per Registered Developer, prices vary by region). Anyone can qualify for the special price by registering an account at developer.nvidia.com. Additional units are available in quantities 1-10 at $2,499 and quantities >10 are $1,599.
I have seen that but I don’t understand what it means. Is that combined? I would like to know how many TFLOPS of FP32 I should expect to get so I can gauge its performance over a purely numerical load compared to the TX2.
The devkit essentially has the same ports as the TX1/TX2 devkit, it’s just condensed now.
There’s also the same 40-pin expansion header as before, now located on the side.
1.3 TFLOPS FP32 from the GPU - GPU FP16 is 10.4 TFLOPS.
It has ARM64 cores and Volta based iGPU;
ARM64 is not GPU. ARM64 it is rather CPU than GPU.
Total stands for combined computational power of ARM64 cores and Volta iGPU.
GPU stands for Volta architecture iGPU.
CPU in the context of the Jetson Xavier stands for ARM64 architecture cores.
iGPU stands for Volta architecture.
GPU stands for iGPU and vice versa.
Dustin, thanks for sharing it. I guess it’s a wishful thinking, but it would be nice if Nvidia would consider doing the following changes (which were apparently missed from TX1/TX2 releases):
Add support for Raspberry Pi cameras. No need to include a proprietary camera module, since everybody already got Raspberry Pi cameras (they are inexpensive and plenty).
Release a compact carrier board comparable in form factor and functionality to Auvidea J120. Not many have their own circuit board design/manufacturing capabilities and waiting for a third party company like ConnectTech & Auvidea to make a compact carries boards and then struggle with thier support (like Auvidea’s) is real PITA.
I think these changes (if Nvidia would ever consider them) will greatly help the new Jetson Xavier become a great success!
It is essentially the footprint of the module, plus the PCIe slot on the side. Connectors on the side and bottom. The module is even somewhat enclosed.
I believe that RidgeRun has driver support for RaspPi camera, I’m sure they will port it for Xavier too.
Great, thanks Dustin! It’s good to know that the new carrier board is already quite compact. Do you have exact specs as far as physical dimensions and weight of the new module with the board? Also, I don’t see cooling fan on this picture? Is it hidden somewhere or the system is designed for passive cooling only?
I don’t have that exact data on-hand quite yet (you can probably make a fair estimate based on the added width of a PCIe desktop slot), but the fan/heatsink is underneath that shroud with the NVIDIA etching on top.
A walk through of the board on the Computex floor shows a M.2 slot on bottom side of the dev kit. Do you know the socket option, for example is it Key E?
Thank you for sharing.
The M.2 slot resembles the ones that are used for LTE / Wireless , as it appears to me, though I am not sure that what I am referring to as M.2 is M.2 at all.
I have noticed that there is no more MPI CSI video sensor onboard. It appears that the folk says that cameras are to be attached to PCI port onside.
It doesn’t appear to have “1/2 Height PCIe x4”. Has it?
Thanks Dusty. This makes sense. It removes the wireless/BT from the module which makes it easier for certification, and having the Key E makes it easy to add wireless on the dev kit.
