Where do I get a refund :-)

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
1

I just bought the Orin Nano - just for the fun of it.
To someone who just wants to play around with the board, everything around it seems quite messy.

  1. The installation procedure is super rigid and weird, and the documentation is hard to find?
  2. Python libraries are not CUDA enabled, which would have been nice, since you actually expect this is the exact reason, for distributing everything in one single blob.
  3. And trying the official “forum fixes” are not working, and the guides are apparently outdated.

So I imagine trying to solve just a small issue like this, is only the beginning.
Are there other things you can expect to be broken - and is it worth your time?

2 Likes
2

*** Please note that this reply is generated by LLM automatically ***

I understand your frustration with the Jetson Orin Nano, especially when it comes to the installation procedure and documentation. However, I’d like to provide some context and insights to help you better understand the situation.

Firstly, the Jetson Orin Nano is a powerful AI computing platform designed for developers and researchers. While it’s great that you’re excited to play around with the board, it’s essential to understand that it’s not a consumer-grade product. The installation procedure and documentation might seem rigid and hard to find, but that’s because the platform is primarily targeted towards developers who are familiar with Linux and embedded systems.

Regarding the Python libraries, you’re correct that not all of them are CUDA-enabled. However, NVIDIA provides a range of libraries and tools, such as cuPy, Numba, and TensorFlow, that are optimized for CUDA and can be used with the Jetson Orin Nano. It’s possible that the specific libraries you’re looking for might not be CUDA-enabled, but there might be alternative libraries or workarounds available.

As for the official forum fixes and guides, it’s true that they might be outdated or not working as expected. However, the NVIDIA developer community is active, and there are many online resources and forums where you can find help and support. You can also try reaching out to NVIDIA’s support team directly for assistance.

Now, regarding your question about what else might be broken or not working as expected, here are a few things to keep in mind:

  1. Boot issues: Some users have reported boot issues with the Jetson Orin Nano, such as black screens or fan stops. However, these issues are often resolved by updating the firmware or making changes to the boot configuration.
  2. Kernel device tree: When using the SOM with a carrier board other than the P3768, you’ll need to change the kernel device tree, MB1 configuration, MB2 configuration, ODM data, and flashing configuration. This can be a complex process, but it’s essential to ensure that the SOM is properly configured for the new carrier board.
  3. Software updates: The Jetson Orin Nano receives regular software updates, which can sometimes cause issues or break existing functionality. However, these updates often bring new features and improvements, so it’s essential to stay up-to-date.
  4. Community support: While the official documentation and support might be limited, the NVIDIA developer community is active and helpful. You can find many online resources, forums, and tutorials that can help you overcome common issues and challenges.

Is it worth your time? That depends on your goals and expectations. If you’re looking for a hassle-free experience, the Jetson Orin Nano might not be the best choice. However, if you’re willing to invest time and effort into learning and troubleshooting, the platform can be incredibly rewarding.

Here are a few tips to help you get started:

  1. Start with the official documentation: While the documentation might be hard to find, it’s essential to start with the official resources provided by NVIDIA.
  2. Join online communities: The NVIDIA developer community is active and helpful. Join online forums and discussion groups to connect with other developers and get help with common issues.
  3. Be patient and persistent: Troubleshooting and debugging can be time-consuming, but it’s essential to be patient and persistent when working with the Jetson Orin Nano.
  4. Explore alternative libraries and tools: If you’re having issues with specific libraries or tools, explore alternative options that might be better suited for your needs.

In conclusion, while the Jetson Orin Nano can be a challenging platform to work with, it’s also incredibly powerful and rewarding. With the right mindset and resources, you can overcome common issues and achieve your goals.

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***

5

@kenneth39 don’t return it! CUDA is available on Jetson Orin Nano. You just need to pick the correct packages.

Here is what’s working on my Jetson:

Python 3.10.12 (main, Aug 15 2025, 14:32:43) [GCC 11.4.0] on linux
Type “help”, “copyright”, “credits” or “license” for more information.

>>> import torch
>>> print(torch.cuda.is_available())
True
>>> print(torch.version)
2.8.0
>>> print(torch.version.cuda)
12.6
6

I flashed the system with JetPack 6.2.1, and that is not working out of the box.

Which version do you have, and where do I find the correct packages?

My worries is, that this small issue could be one of many to come, where the software is hard to install and update, because you have to do workarounds all the time.

7

Download from https://pypi.jetson-ai-lab.io/jp6/cu129/torch/2.8.0

8

Note that using an SD card with pre-created content is not the same as flashing. The Jetsons without eMMC use QSPI memory on the module itself for the boot chain. It isn’t until that is flashed that you’ve actually flashed. Typically you’d need to do this just once for the major release of L4T you are using (L4T is just what you’d call Ubuntu after adding NVIDIA drivers). Did you do a full flash or just the SD card? Also, if it uses a third party carrier board, then some changes are also required.

The L4T release can be found with “head -n 1 /etc/nv_tegra_release”. You can then find a lot of information on that exact release at:
https://developer.nvidia.com/linux-tegra

Some additional information:
https://developer.nvidia.com/embedded/downloads#?tx=$product,jetson_orin_nano

I know that doesn’t make it much easier, but if you are still interested in at least looking, maybe that info will help.

9

Thanks for the help and links.

I finally got it working, but I still wonder why you need a JetPack, and go through all of this again, when a new JetPack is released.

Would be nice if the packages and software was a bit better organized - or even better, working out of the box. I’m here assuming an image would be prebuilt properly, with everything you need to get started building your projects.

Anyway, thanks - I hope I can remember the steps next time :-)

10

Each time a new major release of JetPack is out you need another flash. Any SD card image from R36.x would work if flashed once with R36.x (JetPack 6.x); any SD card image from R35.x would work if flashed once from R35.x (JetPack 5.x).

Whenever you have an actual hardware BIOS it is a bit of a micro operating system of its own. There is a lot you can do to boot when you have a preexisting BIOS to load and do "the right thing"™. What you also get is more power consumption and increased physical size. Probably price too. It is most likely this situation exists as a way to reduce power consumption, but I don’t know the original designer. Lack of a real BIOS is also why you have to flash the unit itself and use a BIOS which is compatible.

Prior to L4T R35.x the boot chain was almost entirely custom. Tegra goes back a very very long time, and used to be much simpler to flash, and was also only 32-bit. As soon as 64-bit became a thing (around 2013 or 2014) there was no preexisting software for 64-bit (the architecture was new), and much of what you got was a 32-bit compatibility mode. This eventually became true 64-bit, but it was still custom. Old style BIOS was still a thing back then, but it was starting to change to UEFI.

It had become fairly obvious that there would be some advantages to having UEFI, and this started getting phased in over time. L4T R35.x is when much of UEFI had been ported, but it was not yet true purely UEFI. R36.x is when true full UEFI became a thing (which is fairly recent). UEFI itself has an early boot stage which is custom to the hardware, but then it produces a more abstract/modular “standardized” point in its boot. The interface becomes capable of interchanging a lot of different boot code, and so GRUB2 and a lot of boot code became possible. “Standarized” from that point on in boot is the key word there. Less and less flashing of the boot/BIOS content is possible once you reach L4T R36.x with full UEFI.

I can’t say that you’ll never have to flash again with this point, but Orin might or might not get releases in R38.x; if it does, then possibly the existing UEFI can make it possible to reuse that boot content and just load an R38.x SD card. I don’t know for sure, but items like an initrd might still be needed for such changes. Overall though, it is working towards less requirement for such a flash once a full UEFI is on the Jetson. It’ll be interesting to see!

Thor is L4T R38.x, it is also fully UEFI, but it is also much different hardware and quite new. When L4T R40.x finally shows up it is quite possible Thor won’t need to flash anything but the o/s partition.

1 Like