I have Jetson Xavier AGX development kit with 512 GB SSD - OS boot drive.I have installed Jetpack 4.6. Today, after installing CUDA 11.0 . It was fine until Ubuntu 18.04 informs me software update (including some from Nvidia). When i install the update, there is some broken installation error… to fix that i have used (with thinking much)
sudo apt-get -o Dpkg ::options::"=--force -overwrite" install --fix-broken
On the same place, some mentioned all the Ubuntu update may not relevant because some not relevant to ARM64. When i reboot the kit, i got the boot issue.
After logo, blank screen -similar to https://forums.developer.nvidia.com/t/jetson-xavier-nx-not-booting/141070
As suggested,i have used to serial able/ Gtkterm login to terminal (default username/ password). From the error message i have found out it was some related to
tegradc and tried to restart
gdm3 service and ‘’ afraid to change anything on ‘/etc/gdm3/custom.conf’. So, install gdm again after reboot still blank screen.
Finally, install the ‘lightdm’ and able to boot to Ubuntu GUI login screen. (Ubuntu Unity desktop environment ) but the can not login with correct username /passowrd . Changed to Ubuntu from Ubuntu Unity. Still login not success (no error msg, but after hitting enter button - did not get the usual Desktop). When choose the Lxde environment, able to login.
From Lxde environment, tr.y to change the display manager through
dpkg-reconfigure lightdm or restart the
gdm3 service leads to blank screen issue after reboot. Also, noticed, on Lxde when i click any icon, it takes slight more time initially otherwise seems all right.
If not mistaken the L4T/ Ubuntu 18.04 on Xavier AGX by default uses Unity desktop environment. How to get that “shiny interface” back with gtm3 enabled?
Let me know, if i missing anything or any question /thoughts highly useful.
As far as I know CUDA, when installed via JetPack/SDK Manager, is still at CUDA 10 (perhaps a preview SDKM has newer? Don’t know). If you downloaded and installed CUDA 11 from a separately downloaded file on a web site, then you’ll probably have to start over…only the CUDA for a Jetson will work with the iGPU (the integrated GPU sits directly on the memory controller, but other CUDA releases expect PCI GPUs…they won’t work when mixing).
Btw, a freshly flashed Jetson has
apt repositories for the same thing which JetPack/SDKM would install. These are quite different than separate download sites you might see for general use. If you have used a force option, then likely you will need to reflash.
I have followed this article for Native boot JetPack 4.6 (L4T 32.6.1) from SSD
 [Native Boot for Jetson Xaviers - JetsonHacks]
Installed CUDA from here:
It seems working fine and no issue. until Ubuntu OS update. After reading your reply, did not understand how CUDA installation affect the display manger. However, still believing the course issue mainly related to GDM display manger. If fixing it, (hopefuly) will solve the issue. Please correct me if i miss some point.
In case of Reflash- Do I have to follow the same method as mentioned . My objective is install OpenCV with CUDA support, Yolo /Darknet on Xavier AGX.
On the JetPack 4.6 page I see CUDA is at version 10.2. If you use someone else’s carrier board, then you’d have to use their board support package, but I don’t see any way that on 4.6 that CUDA 11 is supported in any way. But that does lead to the other question: Is this a third party carrier board? If so, then you’d have to use their device tree for flash, but I see no way to use CUDA 11. Perhaps it is a preview release?
Note: The failure upon update is likely the device tree and/or boot content being updated for the dev kit and not for a custom carrier board (assuming this is not a dev kit).
I am using Jetson AGX developer kit and beginner. When i read CUDA 11 has a wealth of features, i was installed it (CUDA) without think twice and did not JetPack 4.6 documentation. It was mistake. After reading Jetpack 4.6 release page useful and understand it not easy to rollback pack to CUDA 10.2. So, in this tricky situation, installed the cudnn 8.0.2 with following command
sudo dpkg -i libcudnn8_126.96.36.199-1+cuda10.2_arm64.deb
sudo dpkg -i libcudnn8-dev_188.8.131.52-1+cuda10.2_arm64.deb
Not sure, this is good idea? For later it may create issues when i try to build OpenCV with CUDA support? or build Darknet with CUDA /OpenCV support. GCC version = 7.5. Noticed the Jetpack 4.6 has comes with 4.1.1.
Is is there any way to install JetPack again as fresh install. This Xavier Kit used for University research. Wish to install stable software/library beginning itself to avoid libraries/ sdk issues later. Any thoughts /advise will be useful.
If you’ve installed
arm64 components from a non-JetPack/SDKM source, then you might as well try it out. However, I suspect this will fail. Much of the other
arm64 package content you can find “in the wild” is for data center hardware, and not necessarily compatible with the iGPU of the Jetson (which is not based on PCI).
To flash again you’d use JetPack/SDKM. I suggest the most recent release from here:
I have followed the instruction here https://docs.nvidia.com/sdk-manager/install-with-sdkm-jetson/index.html
and installed the Jetpack 4.6 on NVM SSD memory. successfully. Using Ubuntu 18.04 and
The screen shot below. Thanks.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.