NVIDIA Jetson AGX Xavier Developer Kit Unboxing and Demonstration
Articles Blog

NVIDIA Jetson AGX Xavier Developer Kit Unboxing and Demonstration

August 24, 2019

hello it’s Jim from JetsonHacks.com on
today’s show we are looking at the Nvidia Jetson AGX Xavier developer kit NVIDIA was kind enough to send me a
review unit. This is the familiar Jetson box. Let’s open it up. Here’s the Xavier! Let’s see what else is in the box. Okay,
so we have the support guide and the QuickStart guide. Put those aside for a
moment. Here’s the power supply. Plug for the power supply. This appears to be a
USB C to USB A cable. Blue indicates that it’s USB 3.0. Let’s take a closer look at
the Xavier. The footprint of the Xavier is 105 millimeter square, and it is 65
millimeters in height. In comparison let’s take a look at aJetson TX2. The
Jetson TX2 is 170 millimeters square and it’s about 45 to 50 millimeters high
depending on how much at the space you leave underneath the board. The Xavier
weighs 665 grams which is around a pound and a half.
The TX2 was a little bit over a pound. The developer kit consists of two parts.
The top part is the Jetson module itself. The processing components of the Jetson
Xavier module consists of an 8 core NVIDIA Carmel ARM version
8.2 CPU, a 512 core Volta GPU with 64 Tensor Cores, dual deep-learning
accelerators (these are referred to as DLA engines), 16 gigabytes of 256 bit
LPDDR4 memory, 32 gigabytes of emmc memory, a vision accelerator engine,
hardware video encoders and decoders and a hardware audio engine. As usual there’s
more information in the article linked in the description below. The Jetson
module is connected to the carrier board using a 699 pin connector. The carrier
board has the i/o connections. This is the front of the module. This is a USB
3.1 type C connector. This is the USB port that you use to flash the device.
Here’s the power light indicator. 40 pins of GPIO
and a microUSB connector, which is a serial FTDI to USB connection. Here we
have a bi-16 PCIe slot Gen 4 controller. On the back of the device, we have a
micro SD card reader. This can also read UFS cards. HDMI 2.0 Here we have a hybrid connector. It’s USB
A type or eSATA. Gigabit Ethernet. Another USB 3.1 type-c connector and a barrel
connector for the 65 watt power supply, which is included in the kit. And on this
side of the kit we have the power button force recovery and the reset button. Let’s take a look underneath the kit. On
the bottom side of the Dev kit we have a camera connector and a M.2 Key E
connector. On the other side of this carrier board is a M.2 Key M connector.
You must remove the Jetson module to access the other side of the carrier
board. The next step after getting the Xavier out of the box is to flash it
using Jetpack. We have another video coming out that will cover that process
in detail. I’ve just finished the Jetpack installation. We’re ready to run some
demos. The Xavier is hooked up to this HDMI display through the HDMI port and
the keyboard and the mouse are connected via a USB hub. We have just finished
installing all the software and some demos. Let’s take a closer look and
switch over to the Xavier screen. Okay now we’re on the Xavier. Let’s take
a quick look around here. We’ll go to our system settings and look at the details.
Here’s the overview. 15.4 gigabytes of memory, processor ARM v8 times 4. This number is a little surprising because we know that there
are eight cores all together. We’ll go over that in a minute. Tegra Xavier,
64-bit OS. So let’s look at the number of cores that are online. It turns out that
we can configure the number of cores online using the NVP model utility.
Let’s query the current mode. We are currently in mode 15-watt. There are
three main modes, 10 watts, 15 watts, and 30 watts. 15 watts emulates a Jetson TX2.
There are seven different MVP models we can run in the stock configuration. They
control the clock frequencies, the GPU frequency and the number of cores that
are online. Let’s switch over to maximum performance mode. When we open up the
system monitor we see that there are eight CPU cores on line.
Let’s close the settings dialog. Reopen it. Details and now there are eight
cores as we expect. Next to the system monitor we have a GPU activity monitor.
The code for this is available on the Github JetsonHacks account. It’s time
to run the demos! This is a CUDA sample, We’ll start this baby up. We can see that we get almost 60 frames
a second, three hundred and some-odd gigaflops per second. The next demo that
we’ll run is one of the VisionWorks samples. This is a Hough transform. You can see that it’s trying to figure
out where the lines in the road are. Here’s another VisionWorks sample. This
estimates motion flow. I like this one because that’s a little doggies in it. And
our third vision work sample is a video stabilizer. We can see the GPU usage down here. The
CPUs aren’t particularly busy at this point. They’re all around 20% or so so.
Let’s run more than one demo at once. So you can see that none of the frame
rates really changed a whole lot here. Tthere still seems to be quite a bit of
computing room to spare. The GPU is a little bit busier now. I
would guess it’s around 50 percent usage. Let’s take a look at our CPUs, they’re in
the 25 to 30 percent range. But it’s a nice little introduction to
what a visual application might look like on the Xavier. If you liked this
video, give it a thumbs up! If you’d like more exciting content like this please
subscribe. Thanks for watching!

Only registered users can comment.

  1. Please! Give us some Yolo benchmarks! A darknet (c/c++) implementation of YoloV2 gives me 3fps with yolov2-tiny giving me 13fps.

  2. In fairness to the TX1 and TX2 formats if you were to use them in practice you'd replace the carrier board with something like this: http://connecttech.com/product/astro-carrier-for-nvidia-jetson-tx2-tx1/ and get much smaller form factor. If you need the compute horsepower though this thing looks epic.

  3. +JetsonHacks do you recommend without being influenced by Nvidias sayings that the Xavier is worth buying? How well do you think it would work if it was used in something like your MIT Racecar but outside, with a Zed Cam (once updated) do you think it will build better maps without having problems?

  4. You look like prof. charles Xavier…. maybe the guys at Nvidia saw your earlier videos and this name popped up in their mind!! room for thought!

  5. Hi,
    Compared to Jetson TX2, how much is the performance improvement? E.g., running a same deep learning model, such as YOLO, SSD, etc, how faster can we get w.r.t to Jetson TX2.

  6. If you would like a more comprehensive discussion of the architecture and performance (measured against the Jetson TX2) of the AGX Xavier, please see the link in the JetsonHacks article: https://wp.me/p7ZgI9-13x under "Peak Performance".

  7. Great and clear videos as always! Any idea if the Xavier is compatible with this camera setup?

  8. What??? NVIDIA use Ubuntu? why they are not customing their linux os with minimal package needed to run the SDK, and why there is office thing there

  9. i was thinking about getting the 99$ one that is coming out because i have been wanting something like the pi but with a little more juice

  10. Can u pls make a Jetson nano video guide too ? I keep going in circles with it and the online guides. I have flashed the SD card as per the instructions but no examples where installed by default nor I understand how to load them using the SDK manager tool as in the instructions for the nano it says it’s not needed

  11. Could you please tell me where can I see the sample code you used? I'm specially interested in movement detection with "doggies")))) Many thanks in advance)

Leave a Reply

Your email address will not be published. Required fields are marked *