Help with Jetson TX2 startup

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
,
1

I am having trouble getting the Jetson TX2 to show something on the screen. I have tried to boot with an SD with Ubuntu images from the official website version 18, 20 and 22 but none of them works. I also tried looking for youtube videos but most of them already have a Linux OS running or they used SDK manager. It’s hard to get SDK manager since I myself don’t have a Linux computer (I do have VMware but im waiting on the key). A community member told me there is an Linux OS image in the download site but I didn’t see any mention of OS in any of them.

Is there any other way I can try to setup the Jetson TX2?

2

Is this a “Nano” or “NX” version of TX2? There are multiple TX2 devices available. The answer has a lot of dependency on which model you have. There are also commercial modules sold on third party carrier boards (then they would no longer be called a development kit or “dev” kit).

One primary difference on the answer is where your SD card slot is. If the slot is on the module, versus if the SD card slot is on the carrier board will change everything. The firmware for third party carrier boards differ; the firmware for SD slot on carrier board also differs from the firmware if the SD slot is on the module itself.

Note that models with SD card on the carrier board do also have eMMC. Models with the SD card slot on the module itself do not have eMMC. Both models have QSPI memory that would be used differently depending on the SD card slot location.

It is very likely you will have to flash something (even if it is QSPI only) using a Linux host PC. VMware is not normally useful due to configuration issues (VMs are not equivalent in all ways). However, some people do make this work. During flash a Jetson is a custom USB device, and the flash software on the Linux host PC has the “driver” for this. During any flash the Jetson’s USB will disconnect and reconnect, and most VMs will fail to reacquire the USB on reconnect. If you configure the VM to grab a reconnect as well as the starting USB, then it can be made to work (it would be up to the VMware support to suggest how to automatically keep the USB of just the one device upon reconnect; the Jetson itself has no control over that).

I’ll also suggest that if you use a VM you need to be certain the disk space used for the Linux host PC is large enough and it must be type ext4 (you cannot use NTFS or VFAT/FAT32 as the filesystem type).

For reference, JetPack/SDK Manager is a graphical front end to the flash software which runs on the host PC during flash. L4T is what actually gets installed, and this is Ubuntu plus NVIDIA drivers. You would want the most recent release your hardware supports, which is likely R32.x. If you have a third party carrier board, then you must use the software your manufacturer provides (which might actually just be a patch to the NVIDIA software or a statement that the NVIDIA software is all that is needed, but you must consult the third party manufacturer if this is not a dev kit). The list of L4T releases is here:
https://developer.nvidia.com/linux-tegra

Documentation also exists there for that release. I’m guessing you will end up with R32.7.4. There are guides there for specific models.

Also, if you have a newer JetPack/SDK Manager, then you can start it like this to see older releases (which might be needed to see R32.x if you have a newer JetPack):
sdkmanager --archived-versions

3

Hi, thank you for the in-depth information that you have provided.
To answer some of your questions, I have looked up the “Nano” and “NX” models and they don’t look similar to what I have. It is a development kit and the model name it provides is “P2597” with 8GB RAM.

For the SD Card slot, I don’t know the difference between the module or the carrier board, however, the SD card slot is located at the side between the HDMI and ethernet port. To give more information of the specification, it has 32GB eMMC 5.1 flash storage.

I noticed in some videos that the heatsink fan turns on when powering the TX2. When I turn on mine, the fan doesn’t turn on. I don’t know if it is defective or because it is not flashed yet.

Lastly, I managed to create a boot Ubuntu with version 18.04 (Bionic Beaver) and I will see if I can download SDK Manager and establish communication with the TX2. I am new with the TX2 board itself so I apologize if the information that I have provided is confusing.

4

I don’t know which specific carrier board you have, but the original TX2 (not Nano, nor NX) is larger square shape. I’m not actually measuring this, but I’m thinking it was about 10 inches by 10 inches. This is a picture of the full TX2 dev kit:
https://developer.nvidia.com/blog/parallelforall/wp-content/uploads/2017/03/JTX2_Devkit.png

Note the heat sink on that board. The module sits beneath the heat sink and looks like this:
https://developer-blogs.nvidia.com/wp-content/uploads/2017/03/Figure1_TX2-e1488772330657.png

You won’t be able to install the NVIDIA flash software on a Windows system, so this might be getting too far ahead, but that software has flash “targets”. Targets are just the specification for flashing a module plus carrier board combination. It used to only be possible to flash on command line, and you still can, but targets can either be from the “technical” name, or from the common “alias”. The technical name is always a combination of the module model and the carrier board model, and sometimes also adds different “purposes”, e.g., a normal configuration or a 24x7 configuration. Here is an example of the configuration file alias and what it points to:

jetson-tx2-devkit.conf -> p2597-0000+p3310-1000.conf

Notice that p2597? It matches your model. That is a module model. The carrier board for that is the p3310, with a revision. The content for a p2597 would be the same for a number of TX2 models, but the p3310 would vary. The result would be different firmware/device tree content.

Does your TX2 look like the larger square model in the first image? If so, then your flash target is likely “jetson-tx2-devkit”.

The SD card slot on that model is not under the heat sink, attached to the module. Instead, the SD slot is along the edge of the carrier board the heat sink unit attaches to. This model has eMMC memory, and differs such that it will not boot to the SD card. For that model you must flash the eMMC. There are more complicated ways to pick other boot targets (such as SD card or USB SATA drive), but this requires extra steps and is more complicated than just flashing to the eMMC. Do not expect an SD card to boot on such a model. No SD card is required for that model.

Jetsons are very power efficient. The fan only goes on when warm enough to need cooling. Most Jetsons won’t power the fan except for a very brief “blip” right at power on, and again only when under load. The fan is not on because (A) it is not flashed, and (B) it is not under sufficient load to heat up.

5

The board that I have is the first image. So the only way to flash the eMMC is through SDK Manager?

6

I managed to download SDK manager and I am now in the process of installing.

In step 2, it shows me the folder paths for “host” and “target”. Obviously the “target” is the TX2 module so it should be installed in the storage of the TX2 itself. By default the path folder is on my Ubuntu storage which I find it strange.

Where should the “download folder” and “Target HW image folder” path be located? Can I install both of these in the same external hard drive?

7

They are on your host PC. Default location is ~/nvidia.

It could be chosen to other drive on your host PC.

8

During the process of downloading the required downloadable contents in SDK Manager, I encountered “Dependency error”.

Below is a error log that terminal show:

11:19:39 ERROR: CUDA Cross Compile Package on Host - host: Download request timed out after 60 seconds. Closing connections
11:19:39 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
11:19:39 ERROR: CUDA Cross Compile Package on Host - host: Download request timed out after 60 seconds. Closing connections
11:19:39 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
11:19:39 ERROR: CUDA Cross Compile Package on Host - host: download failed
12:57:43 SUMMARY: CUDA Cross Compile Package on Host - host: Depends on failed component
13:02:52 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
13:02:52 ERROR: CUDA Cross Compile Package on Host - host: token_rejected
13:03:16 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
11:19:32 ERROR: VPI on Target - target: Download request timed out after 60 seconds. Closing connections
13:03:16 ERROR: CUDA Cross Compile Package on Host - host: token_rejected
13:03:44 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
13:03:44 ERROR: CUDA Cross Compile Package on Host - host: token_rejected
13:04:21 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
13:04:21 ERROR: CUDA Cross Compile Package on Host - host: token_rejected
13:05:13 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
13:05:13 ERROR: CUDA Cross Compile Package on Host - host: token_rejected
13:05:13 SUMMARY: CUDA Cross Compile Package on Host - host: Depends on failed component
13:05:13 ERROR: CUDA Cross Compile Package on Host - host: Download ‘CUDA Cross Compile Package on Host’ failure
13:05:13 ERROR: CUDA Cross Compile Package on Host - host: download failed

Does anyone know how to fix this?

9

I managed to fix the issue above by doing the following command:
$ sudo apt-get update
$ sudo apt-get -f install

Source: SDK Manager missing dependencies

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.