NVIDIA Jetson TX2 Setup

If your TX2, like mine, came without an operating system, you will have no luck booting into a native Linux environment. In any case, NVIDIA suggests updating JetPack to the latest version.

So let's get a Linux environment, JetPack, Cuda, and a few other items installed.

JetPack 3.1 Installation

First of all, much credit is owed to Klein Yuan on Github for his tutorial. This walk through will build off of that. Also, here is an awesome walkthrough on Youtube of the setup and install of JetPack, but some of the details were missing for a VM (hence why I wrote this!)

I will be using a MacBook Pro(running 10.12.6, Sierra) with VirtualBox to flash the TX2

The entire process takes about 2 hours(depends on connection/data transfer rate)

1. Get VirtualBox to run a Linux Environment

Since the Jetson TX2 must be flashed with a Linux environment, and ideally 14.04 Ubuntu(have heard of 16.06 working also), we will need to create a VM with Linux running.

Download VirtualBox here, or use VMWare if you're on Windows. With VirtualBox, you're also going to need the Extension Package to use USB 2.0/3.0. This can be directly downloaded here

Once you have VirtualBox, we need to get a boot image for Linux.

2. Download Linux Boot Image

Since we need to use Linux Ubuntu 14.04, let's grab that from the Linux Downloads page.

This link will directly download the .iso file for 64-bit desktop machines.

3. Setup VM

With the Linux Boot Image downloaded, we need to setup the VM and boot into it.

Select the New Icon wheel in the upper left of VirtualBox. Fill in the necessary information. You should select at least 1+ GB RAM, and 32GB HD.

Go to Settings --> Network --> Adapter 1, change Attached to to Bridged Adapter, and name en0: Wi-Fi

Now launch the VM you just created. At this point, if it hasn't already, VirtualBox will ask you which boot image it should use. Navigate to your .iso file, and select it.

4. Setup Linux

Once the Linux environment comes up, it will ask if you want to try out Linux, or do a full install. It worked for me to do a full installation. Select that option, and walk through the Linux Install.

5. Get JetPack

Once inside the Linux Environment, get NVIDIA JetPack here (at the time of this writing it was version 3.1). You're going to need to setup a developer account with them, if you don't have one already!

Once this is downloaded, take the file and put it in a directory like ~/documents/JetPack/

6. Wire the NVIDIA Jetson TX2

With the Jetson shutdown and unplugged from power, plug in the:

• HDMI port to a monitor
• Ethernet port to a router you are connected to on your Mac(can be wired or wireless - I used wireless, but ran into a small problem which I provide a solution to later on)
• Micro USB port with the provided USB cable, and connect that to your Mac - which we call the Host
• USB port to a hub(ideally) with mouse and keyboard. If not, just to a mouse or keyword and switch when needed
• Antennas that came with the Jetson
• Finally, the power cable

7. Put the NVIDIA Jetson TX2 into Force Recovery Mode

From idle mode(with a single red LED illuminated on the Jetson), put the Jetson into Forced recovery mode, by following these steps:

• Press and release the Power Button (furthest to the Right, of the 4 buttons onboard)
• Hold down the Forced Recovery button (says 'REC' on the pcb)
• With REC held down, press and release the reset button (says 'RST' on the pcb)
• Hold REC for 2 more seconds, and release

To verify you've done this correctly, open up a terminal and list the usb devices with one of these commands:

$ ioreg -p IOUSB 
$ ioreg -p IOUSB -w0 -l
$ ioreg -p IOUSB -w0 | sed 's/[^o]*o //; s/@.*$//' | grep -v '^Root.*'

You should see NVIDIA Corp. APX

8. Add NVIDIA to VirtualBox USB Settings

Now, with your VM shutdown, open up the settings of your Linux VM:

Go to Settings --> Ports --> USB, Enable USB Controller - USB 3.0 (xHCI) Controller

Select the plus with usb icon on the right, select NVIDIA Corp. APX

Now, launch the VM, and open a new terminal to verify these settings are functioning and the VM can 'see' the Jetson. Type:

$ lsusb

You should see NVIDIA listed.

9. Run and Install JetPack

Now, on the Linux VM, open a new terminal and set the permissions to executable for the file we downloaded earlier:

$ cd ~/documents/JetPack/
$ chmod +x ./JetPack-L4T-3.1-linux-x64.run

Run jetpack:

$ sudo ./JetPack-L4T-3.1-linux-x64.run

This should bring up a GUI. Click through, select the TX2, do a full installation, accept the terms, and click next.

This will download a lot of items onto the VM, and take a bit of time. At one point you will see this screen:

Since we have already put the device into forced recovery mode, we can open a new terminal, verify we can still 'see' the NVIDIA:

$ lsusb

Then, navigate back to the window, and press ENTER, as it indicates. The installation will continue, and eventually you will see some activity on the monitor connected to your Jetson.

If everything went well, a Linux Ubuntu desktop should launch.

10. Troubleshooting

To test that you received all of the packages with the Linux environment, try out a sample. Navigate to this directory(if you're like me, you won't even have a CUDA directory installed!):

/home/ubuntu/NVIDIA_CUDA-<version>_Samples/bin/armv7l/linux/release/gnueabihf/

This rendering of water will be shown:

So for the rest of us who might have had issues with finding the IP address for the target...

Re-launch jetpack:

$ sudo ./JetPack-L4T-3.1-linux-x64.run

Click through again, but when you get to this screen:

Select Custom install in the upper right corner. And click Clear Actions.

Now, go down the list and click on the Action, and select those in the lower part of the list for the Target. Make sure you deselect Flash OS. Click next.

It will then ask you for your IP address:

To retrieve this, open up a terminal on the Linux VM, and type:

$ hostname -I

Type this 192.XXX.X.XX string into the first box, and your credentials into the next two; which by default are both nvidia.

Click next, and you should be good to go!

Go back to the top of this section, Troubleshooting, and try the water samples and a few others to be sure you have Cuda!

Additional Resources

If you ever have to reflash the target, you can do so directly with:

$ sudo ./flash.sh jetson-tx2 mmcblk0p1

Here are a few other guides from NVIDIA, albeit somewhat limited: http://docs.nvidia.com/jetpack-l4t/2_1/content/developertools/mobile/jetpack/jetpack_l4t/2.0/jetpack_l4t_install.htm http://developer2.download.nvidia.com/embedded/L4T/r28_Release_v1.0/BSP/l4t_quick_start_guide.txt