Hi,
I am testing the basic CUDA example code on the Jetson Orin Nano SD card image, which was downloaded from the Jetpack 6.0 production release link. This prebuilt SD card image includes CUDA 12.2, and our test example CUDA file “cuda_hello_world.cu” runs without any errors.
https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v3.0/jp60-orin-nano-sd-card-image.zip
cuda_hello_world.txt (971 Bytes)
Subsequently, we flashed the sample root filesystem from the given link and using the following command: sudo ./flash.sh jetson-orin-nano-devkit mmcblk0p1. Then installed the CUDA 12.2 package on the Orin Nano developer kit
https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v3.0/release/tegra_linux_sample-root-filesystem_r36.3.0_aarch64.tbz2
With the sample root filesystem, executing “cuda_hello_world.cu” results in a segmentation fault. Are there any driver or package dependencies required for running CUDA 12.2 on the L4T_36_3 BSP kernel?
The nvidia-smi command does not print the CUDA version, and the display screen turns black immediately.
Hi,
Have you installed the CUDA library after flashing?sudo apt-get install nvidia-cuda.
Thanks.
Hi,
CUDA Toolkit 12.2 Downloads | NVIDIA Developer
Executed the below commands to install the cuda toolkit 12.2.
And also tried with sudo apt-get install nvidia-cuda and it didn’t resolve the issue.
The jetpack 6.0 production release runs with 22.04 and the download page doesn’t show the compatible version for 22.04.
Hi,
The simplest way is to install it with our apt command:
$ sudo apt-get update
$ sudo apt-get install nvidia-jetpack
Could you check if this fixes the issue?
Thanks.
Hi,
Tried the above-mentioned steps and still it didn’t resolve the issue.
Can you please follow the below steps and try to reproduce the same error at your end.
Download and Extract the NVIDIA Jetson Linux 36.3 BSP Driver Package
https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v3.0/release/jetson_linux_r36.3.0_aarch64.tbz2
Download and Extract the sample root file system
cd /Linux_for_Tegra/rootfs
https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v3.0/release/tegra_linux_sample-root-filesystem_r36.3.0_aarch64.tbz2
Flash the image with custom kernel and complete the installation process.
sudo ./flash.sh jetson-orin-nano-devkit mmcblk0p1
Try to run the below command and test the example cuda program
cuda_hello_world.txt (971 Bytes)
$ sudo apt-get update
$ sudo apt-get install nvidia-jetpack
$ nvidia-smi
$ nvcc cuda_hello_world.cu -o cuda_hello_world
$ ./cuda_hello_world
Thanks
Hi,
Thanks. We will give it a try.
Thanks.
Hi,
Yes, we have tried the below CUDA installation commands, and it didn’t resolve the issue. After running the example program, it gives segmentation fault.
$ sudo apt-get update
I have updated the CUDA installation command in the previous comment and please check it once.
Thanks
Hi,
Could you reflash the device and try it again?
$ ./cuda_hello_world
Max error: 0
$ /usr/local/cuda-12.2/bin/nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:08:11_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0
Thanks.
Hi,
yes, the same steps are working with my setup. This uses default kernel available in the /Linux_for_Tegra/kernel/Image.
When I overwrite the custom kernel and kernel modules build from the L4T_36_3 BSP source, the issue is reproducible.
Can you please share the defconfig file of the default kernel or list down the other drivers’ dependencies needs to be added for the custom kernel.
Thanks
Hi,
Do you follow our document below to build the custom kernel?
https://docs.nvidia.com/jetson/archives/r36.3/DeveloperGuide/SD/Kernel/BringYourOwnKernel.html
The defconfig can be found in the source link provided above.
Thanks.
system
August 28, 2024, 7:48am
13
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.
