Sorry, I somehow forgot a key piece for getting GPU programs running inside of Docker containers.
Need to set the
LD_LIBRARY_PATH
to point to the
/usr/lib/aarch64-linux-gnu-tegra
directory inside of your Dockerfile as it contains libraries that your GPU program will need to run. There might be others as well depending on your specific program, but for deviceQuery, it needs that directory specifically.
I’ve updated the README as well as the sample Dockerfile to reflect this.
I am a student working on Docker project on Nvidia jetson Tx2, I read all the comments on this topic. I tried installing Docker normally as given on official docker documentation. But got to know Docker does not support for Nvidia tegra devices and now after seeing the info here I am hoping to get docker run on my jetson Tx2.
My whole project is to implement OpenCV app on Jetson Tx2 by using GPU capababilty and I should dockerize the Application. For this I need to first install docker on Tx2 as you guys explained in the Github links. Next I need to give containers access to use GPU feature. Later on I should be able to provide High availability of this app by pushing the app to registry Dockerhub.
So I thought the above guide provided by you all pretty much satisfies all the requirements to attain the objective of project.
Shall i proceed with the steps as you all suggested? Can we manage swarms on jetson Tx2 by creating clusters and deployment of nodes across multiple hosts as we do in Docker on normal PC’s ?
It would be really helpful if you reach out to me via email: nagthejmr@gmail.com
Can we manage swarms on jetson Tx2 by creating clusters and deployment of nodes across multiple hosts as we do in Docker on normal PC’s ?
We don’t have a relevant experiment before. But ideally, it should work.
Has anyone installed Nvidia-docker on jetson Tx2?
I saw we can make use of Containers and work like a normal docker and leverage GPU of jetson. But i am not able to find the exact link of the installation guide.
If anyone is aware of this, please provide provide me a source link to follow instructions to install it.
Nvidia-docker is not supported TX2 right now. (possible in nvdocker2.0)
But you can check this issue; someone has posted an alternative to install Nvidia-docker.
Correct, nvidia-docker is not supported on the TX-2 for a variety of reasons. Based on feedback from NVIDIA they will look into supporting it once nvidia-docker 2.0 is released but it’s unclear exactly where it is on their roadmap.
There are alternatives in the short term which you can view in this thread.
Regarding recompiling the kernel for Docker. I am able to run through the stock recompile without issues but when I replace the .config in /usr/src/kernel/kernel-4.4 with buildJetsonTX2Kernel/config at master · frankjoshua/buildJetsonTX2Kernel · GitHub the build asks a bunch of Y/N questions and doesn’t end up building the arch Image file that copyImage.sh is looking for.
The build goes through all the way from what it looks like but doesn’t create the Image file
I created and Ansible playbook to install Docker. It downloads the kernel sources, copies the config, builds the kernel, reboots, and then installs Docker.
The latest version of Jetpack, 3.2, has support for Docker out of the box. Just need to reflash your TX2 with the latest version and you should be good to go.