docker jetson-nano

[jap937]

a relatively easy way of building a cuda enabled docker is explained recent blog* . However I require the newer L4T 32.4.1 (JetPack 4.4).
I am currently working on a project requiring deepstream which require cuda 10.2
are there any plans for newer releases. or can it be build with the newer nvidia-l4t-core.

*https://www.balena.io/blog/getting-started-with-the-nvidia-jetson-nano-using-balena/

[gelbal]

Hey @jap937, first of all welcome to balena forums!

Looking at the L4T archive I don’t see v32.4.1 released. Then balenaOS v2.56.0 is based on L4T 32.4.2 and the newer balenaOS v2.60.1 (on staging right now) is based on L4T 32.4.3.

CUDA 10.2 is listed as supported on L4T’s website here for version 32.4.2: https://developer.nvidia.com/embedded/linux-tegra-r32.4.2

So for now can you try balenaOS v2.56.0 and see how it goes?

I pinged my teammate who wrote the blog post to see if he has more comments.

[jap937]

thanks for the answer and let me rephrase that i require L4T 32.4.x
this indeed confuses me a little since balenaOS is based on 32.4.2, however the base image "balenalib/jetson-nano-ubuntu:bionic " seems to be based on L4T 32.3.1

[acostach]

Hi @jap937,

Yes, the base images will be updated as soon as all jetson boards get in production to 32.4.x

You can edit /etc/apt/sources.list.d/nvidia.list in the container to switch to r32.4 before installing the packages you need.

[jap937]

thanks, this makes installing jetpack much easier

[pipex]

Great to hear that worked. Let us know if you have any more questions.

[Clint]

Hi,

I don’t understand how this should work, the packages installed are the latest (32.4.4) while balenaOS is still based on 32.4.2 . How do you manage the patch version?

Kind regards,
Clint

[jap937]

the blog mentions that a small difference in OS will most likely work with possible minor glitches.

i did just now notices that i accidentally posted the old jetson guide:
link https://www.balena.io/blog/how-to-use-nvidia-jetson-devices-on-balenaos/

[Clint]

Ok thanks, but our goal is no glitches. We made our own base image from NVIDIA sources in another way. I can share it tomorrow, if you’re interested.

[jap937]

I am very much interested. I am attempting to build one myself as well, but this method turns out to lack certain dependencies for gstreamer

nvidia gstreamer jetson

hello, I am attempting to get my gstreamer application working on the jetson nano with Jetpack 4.4, but i run into a segmentation fault when inspecting the nvvidconv plugin. any ideas on the cause? Dockerfile: FROM balenalib/jetson-nano-ubuntu:bionic RUN echo 'install jetpack4.4' &&\ sed -i 's/r32 main/r32.4 main/g' /etc/apt/sources.list.d/nvidia.list &&\ apt-get update &&\ apt-get install -y nvidia-l4t-core &&\ echo "/usr/lib/aarch64-linux-gnu/tegra" >> /etc/ld.so.co…

[anujdeshpande]

Hi

If you look at that thread again, one of my colleagues has replied how to fix the dependencies for gstreamer. Did that work for you? If not can you reply with your logs on that thread so my colleague can take a look again

[Clint]

Following repo shows how to create working Docker images:

citymesh/jetson-nano-images

[jap937]

thanks, this seems extremely helpful, i already see some things i could do differently

[Clint]

Good to hear it’s helping you. If you have some improvements, you’re welcome to contribute.

[jap937]

im wondering, when you run $ gst-inspect-1.0 nvconv
do you also get ‘segmentation error(core dumped)’?

[gelbal]

Hey @jap937, I see that you asked the same question on this thread and with more details:

nvidia gstreamer jetson

i upgraded everything to 32.4.4 but ‘gst-inspect-1.0 nvvidconv’ still results in a segmentation error .......lots of plugin properties bl-output : Blocklinear output, applicable only for memory:NVMM NV12 format output buffer flags: readable, writable Boolean. Default: true Segmentation fault (core dumped) bottom : Pixels to crop at bottom flags: readable, writable Integer…

If that’s fine with you, I’ll kindly suggest to keep the discussion going in this linked thread. Simply for the sake of tidying up to keep the discussion focused in one location.