I’m to install and use Jetpack 6 on a Jetson Orin Nano 4GB module with the DSBOARD-ORNX from Forecr, however, I am encountering CUDA driver issues that make the board unusable.
In short, after following the flashing and SDK install instruction on the Forecr website, the board will boot fine but any attempts to interact with the GPU will be met with an error and kill the display (the device can still be accessed through ssh).
Please find my system info from jtop here:
Example errors:
Run nvidia-smi:
(I understand that Jetson is an igpu, however I was not expecting a complete failure)
Running the deviceQuery utility from cuda-samples:
Attempt to run an l4t container with the nvidia runtime (e.g. sudo docker run -it –-privileged –-runtime nvidia nvcr.io/nvidia/l4t-jetpack:r36.3.0) “ NVrmMemInitNvmap failed with Permission denied
356: Memory Manager Not supported
NvRmMemInit failed error type: 196626 ”
Debugging Attempts
Here is a list of all the debugging steps I have attempted which resulted in the exact same error:
Rebooting
Re-flashing with fresh install.
Installing SDK components manually via terminal after flashing.
Substituting SSD and flashing again.
Substituting Orin Nano module with another and flashing again.
I have confirmed that the board and SSD are operating normally as Jetpack 5 can be flashed, installed and run perfectly fine following the Forecr guide.
I have also been in contact with @mehmetdeniz at Forcecr support to fix this issue. Currently, it is unclear if the issue is BSP-related or not.
Any help that you could provide would be much appreciated. Thank you!
When you set up the device with the Forcecr guide, have you selected the correct module?
Open the NVIDIA SDK Manager. Select the correct JetPack version for Target Operating System and select the correct module for your installation (“Jetson Orin Nano modules” or “Jetson Orin NX modules”).
Yes, I have. When plugged in through the flashing port the target hardware is recognised as “Jetson Orin Nano 4GB” and the resultant folder under ~/nvidia/nvidia-sdk/ is “JetPack_6.0_Linux_JETSON_ORIN_NANO_TARGETS”.
I have also made sure to check that I have downloaded the correct BSP (“orin_nano_JP6_0_bsp.tar.xz”).
Do you get any feedback from the custom carrier board vendor?
Could you try to connect the module with the Orin Nano carrier board with JetPack 6.0 to see if it can work?
Yes, I’ve been in contact with @mehmetdeniz from Forecr where I described the same errors and debug processes as I’ve done here. The outcome was “Unfortunately, this problem is out of BSP. It’s related to the CUDA and the Docker packages. The NVIDIA moderators will help more efficient instead of us.”
“Could you try to connect the module with the Orin Nano carrier board with JetPack 6.0 to see if it can work?”
I don’t understand what you mean by this question; the Orin Nano module is already connected to the Forecr carrier board and running JetPack 6.0. I can access the desktop environment perfectly fine; he issue is when I attempt to run anything CUDA related.
We found the root cause of this issue. We removed the OVERLAY_DTB_FILE patch from our BSP package and reflashed our Jetson. We could be able to passed the deviceQuery test with it.
Could you please reflash it with the latest BSP files regarding the installation guide?
