I am a Newb. Been playing with raspberry pi’s, intel nucs with OMV on and been getting well away with my home servers. But setting this Jetson up to just do the basic things is terrible. Comes with Jetpack 6.2 preinstalled. So used the 36.4 image for the sd card. Then migrated it to the NVME ssd 2 T. Boots no problem, set up remote desktop etc. Installed the jetson-containers as per the instructions. But why dont these jetcon containers stay resident when you reboot. Also Chrome and Firefox does not work on the distro. Reflashed and started a few times already, managed to get docker and ollama running with llama3,2 , have openwebui running on the intel nuc . Worked for a while and then crashed. The ubuntu just ssem to suc on this and all the instruction manual;s and assisted setup programs just confuses it even more, never mind myself. So if someone has an up to date walkthrough of setting this up, please let me know. Tx.
Hello,
Thanks for visiting the NVIDIA Developer forums.
Your topic will be best served in the Jetson category, I have moved this post for better visibility.
Cheers,
Tom
Hello @rudi.bester,
Sorry to hear you are having a bad time with setting up your Orin Nano.
Just to make sure we are on the same page, let me asl you a couple of questions:
- What is your use case?
- What libraries do you need on the board?
- What is right now stopping you to continue working on your project?
best regards,
Andrew
Embedded Software Engineer at ProventusNova
The browser issue is caused by the snapd. See details at Why Chromium Suddenly Broke on Jetson Orin (and How to Bring It Back) - JetsonHacks
The easy fix from the above link is:
snap download snapd --revision=24724 sudo snap ack snapd_24724.assert sudo snap install snapd_24724.snap sudo snap refresh --hold snapd I applied it to a Jetson Nano and downgrading the snapd fixed the issue of lauching a browser.
I’m also a beginner and encountering the same issue with both Chrome and Firefox; this problem is very troubling for me. Thank you for providing browser troubleshooting information.
Hii Andrew, tx for the reply. As i am still starting to dissect what i can do and cannot do with the jetson, i for the moment just want a LLM Ollama running in a docker container. I already have openwebui running on a IntelNuc with docker on that unit. I had it working about a week and a half ago asking and probing the LLM and then when i rebooted it, the ollama Container was gone.
But what i would like to have and what i am working towards is, ollama running in a container and if possible the visual and text generation in a seperate docker instances where openwebui can use it. Does that make sense. But basically i am playing around now to see what it is capable.
But i would first love to have a stable distro running and not thing disappearing every time i reboot the unit. So learn as i go. Any advice is always welcome.
Tx.
i am using a remote connection from a windows desktop, so for now that does not bother me. But after all the updates in Ubuntu linux, one would think they have sorted out the snapd issue. But hey, we will learn together.
I have learnt.
- Flash img to sd (new how to do that)
- Start unit, no updates yet.
- Install Gparted.
- Migrate sd to NVME
- Reboot into NVME ssd.
- Use Gparted to expand partition for more space.
- Install remote desktop
- Move Unit to location with Lan cable.
- Remote connect and then do updates and upgrades
- Install Portainer to manage containers visually, CLI s bit difficult.
- Then Ollama
- Then explore next.
So let me know what you are doing. Maybe we can learn from each other.
Hi Tx. But decided to do remote connection to windows PC and do all browser stuff from there. Would think by now someone would have fixed the snapd packages by now with the upgrades.
There’s a simpler option to install to SSD. After first boot using the SD card download the latest image and use dd command to write it to the SSD. Remove the SD, reset power and finish the setup. It will ask how much user space to use, select 0 and this option will use the maximum space available on the SSD, gparted is not needed.
For burning the SD card rufus will work fine. balenaetcher is an Electron based app, and bulky!
Hey @ryomahan1996,
Thanks for sharing further details about your project.
Sounds incredibly interesting.
For a more stable Docker setup, maybe you can consider using Docker Compose and have your container restart always. That way you ensure that if the container crashes it will restart, and you would be able to check the container logs with docker logs <container-name>.
best regards,
Andrew
Embedded Software Engineer at ProventusNova
Hi,
Thank you for the information about DD, i will definitely start using that in the future if i will be working it from the command line after playing around and getting to know the unit better/
Rudi
Hi Andrew,
Thank you. I was thinking if how it works, but had nightmares about this last night. My brain kept me awake saying the whole time i am missing something running it that way.
I feel that using it that way i cant use the Jetson packages. Example. I installed portainer and set up a network that i want to use for my containers. Then got ollama/ollama image with NVIDIA settings and then also activated the GPU in the host settings and container settings.
It runs , llama3.2 very quick, but deepseek takes a bit longer due to all the thinking. But then i read this does not have CUDA working in a normal docker. Does CUDA not make it work faster. So i got Dustynv/ollama 36.2.0 and something and installed that. But now i found it does not use the IP4 static ip i assigned it and cant access it now in my remote Openwebui. And this is what bothers me, jetson containers can only be run via the jetson=container cli. Was thinking i can add the –run_unless_stopped command (not sure if that is it, but will find it via google) with a Specific IP4 addess and Host Mode.
So is CUDA not active in a normal docker compose environment and how much performance do i loose by using it this way.?
Thanks for the help.
Rudi
Hello @rudi.bester,
I don’t think using docker compose or not should have any effect on what packages you will be able to use on your Docker container. It is just a tool for setting up and launching containers in an organized and (in my opinion) easier to maintain way.
The packages you have available on your container will depend solely on the Docker Image you use and how you setup your Dockerfile.
Please let me know if that addresses your question.
I am not sure I understood fully your concern.
best regards,
Andrew
Embedded Software Engineer at ProventusNova