Problem with Cuda Toolkit and kernel headers on Jetson Nano SeeedStudio reComputerJ1010

Hello community,

My card was delivered with built-in flash memory and a pre-flashed OS (I didn’t use an SD card for booting).
I would like to use numba and the performance of my graphics card. However, I no longer have all the CUDA Toolkit (CT) on my card (I’ve had to delete some APIs to save space).
So I want to install CT on my embedded board (Jetson Nano, Ubuntu 18.04, aarch64).
Looking at the NVIDIA repository, I see that the range is only compatible from version CT11.8 for my board. But I already have cuda-10.2 installed on my card. I find this strange. Is there any way of reinstalling CT 10.2 (Ubuntu18.04 and aarch64/arm64) from a link? Or can someone explain to me why cuda 10.2 is installed?
I don’t want to install version 11.8 or higher because it would require Ubuntu20.04, which I don’t have.

As a prerequisite, to install CT, you need kernel headers and up-to-date drivers, but I don’t have either (“sudo apt-get install linux-headers-$(uname -r)” doesn’t work).
Also, editing the “nvidia-l4t-source.list” file in the “/etc/apt/source.list.d” directory, then “sudo apt update” “sudo apt upgrade” changes nothing. I even downloaded the file “nvidia-l4t-kernel_4.9.299-tegra-32.7.3-20221122092935_arm64.deb” but I’m afraid that running a dpkg command to install it won’t solve the problem. Should I try?
I’ve seen that you can use SDK Manager but I don’t have an Ubuntu host.

Thank you in advance for your reply and your help.

Hi,

For Nano, you can only install CUDA from JetPack (SDK manager).
And the latest version is 10.2.

Thanks.

Hi AastaLLL,
Thanks for your reply.

Is it the only way ? I have seen that you could directy download Jetpack from the NVIDIA repo.

[how-to] Jetson nano, updating jetpack from 4.4.1 to 5.0.2 - Jetson & Embedded Systems / Jetson Nano - NVIDIA Developer Forums

Index (nvidia.com)

As I have an integrated eMMC module, is it also possible to use SDK manager?

Best regards,
nono_o

As mentioned previously, to install CUDA via SDK manager is the only way.
That link is not suggested by us.

For Nano eMMC module, yes, please use the SDK manager.

Thanks

Thanks for your reply but I am still confused.
Why on the Cuda toolkit10.2’s repo: CUDA Toolkit 10.2 Download | NVIDIA Developer the jetson aarch64 is not included ? Moreover, I don’t have Ubuntu but CentOS 7.7.
So, for my case, which architecture do I use ? Also will the .deb will work ?
Regards,
nono_o

Hi,

For Nano, you can only install CUDA with JetPack.
JetPack can be found either in SDKManager or the online link:
https://repo.download.nvidia.com/jetson/

We do support Jetson on the CUDA download page from CUDA 12.
But it needs JetPack 5 which means Nano is not supported.

Thanks.

Thanks for your reply.
As I’m forced to use sdk manager, I’m going to do it, but for security reasons I’m not going to connect to the Internet. So all I have to do is download “nvidia-jetpack_4.6.1-b110_arm64.deb” on my system. Does this file include the things I’m missing (kernel headers? cuda toolkit?) because I also see “nvidia-l4t-kernel-headers_4.9.299-tegra-32.7.3-20221122092935_arm64. deb” and “nvidia-l4t-cuda_32.7.3-20221122092935_arm64.deb” or “nvidia-cuda_4.6.3-b17_arm64.deb” (what’s the difference between these two?).

Finally, just to be sure, the files are .deb but I’m using a CentOS as host computer, will it still work or do I have to look for the files from another repository?

Best regards,
nono_o

Hello again,
So I m struggling with sdkmanager and the flashing operation. I have SDKManager1.8.0.10363. First I used Ubuntu22.04 but this OS was not supported by the Jetson Nano because on the “target operating system” step, nothing was available for this version. I tried to install Ubuntu 18.04 on 2 different computers, but as the computers were too new, they did not support this OS. So I managed to find a CentOS.7.9.1908 but once again I have the same first issue. So my questions are : what OS on my host computer should I get for my Jetson Nano? Is there, instead, a version of sdkmanager that can fix my issue ?
Best regards,
nono_o

Hi,

You can try the SDKmanager container:

Thanks.

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