On 6 August, we’ll get the first Focal Fossa (20.04) point release. This is generally considered the time when older LTS’s are “safe” to upgrade, and is when users will be prompted to upgrade.
What is Nvidia’s schedule for moving Jetpack over to 20.04?
I know that JP 4.3 works with FF 20.04 great.
Hi vegabook,
That’s under planning, but no schedule yet. At least it won’t happen at this year.
how do you install it?
At least to me getting Ubuntu 20.04 running on Jetson devices is critical. We rely heavily on ROS and that means as long as Ubuntu 20.04 is not supported we are stuck with python2. But I can’t get tensorflow working with python2 on the Jetson devices, so this is actually making development on these devices incredible hard.
I have no experience with your software mix, but FYI, Ubuntu 18.04 has available both Python 2 and 3. The default may not have what you want, but to illustrate, what do you see from:
which python
which python2
which python3
apt search python | egrep '^(python[23][.][0-9]-minimal)'
The package might already be installed, but the path used to specify might need to specify which release. Or the package might not be installed, and getting it would be as simple as installing the other release (leaving the old one still installed), and then specifying to use python3.
Yeah, python2 works just fine, so does python3. It just means all packages relaying on ROS need to stay compatible with python2, which is a development burden, also I haven’t found any recent tensorflow weels for python2 so can not use tensorflow anywhere. It would be really great if we could move everything to python3, which is possible with ROS Noetic on Ubuntu 20.04.
Hi @cosama, you can compile ROS Noetic from source for Ubuntu 18.04.
I recently made a Dockerfile for building ROS Foxy from source for Ubuntu 18.04 here - https://github.com/dusty-nv/jetson-containers/blob/master/Dockerfile.ros.foxy
I will make one to build ROS Noetic as well and let you know when it’s available.
@dusty_nv Thanks. Building from source is a great idea, didn’t thought of that. Happy to try out the docker once available.
I am running 4.4 on Fossa.
Clean install on Beaver then update then upgrade to Fossa
Here is a Dockerfile for building ROS Noetic from source for 18.04 on JetPack 4.4: https://github.com/dusty-nv/jetson-containers/blob/master/Dockerfile.ros.noetic
See the readme from GitHub - dusty-nv/jetson-containers: Machine Learning Containers for NVIDIA Jetson and JetPack-L4T for instructions on building the containers.
Is there any update on the timeline for Jetpack moving to Ubuntu 20.04? It would be great to understand this timeline to support migration planning.
No, I haven’t seen the plan yet.
I’d like to see Jetpack on Ubuntu 20.04 too. Doesn’t that also require TensorRT to support Python 3.8 too, since Ubuntu 20.04 uses Python 3.8? I tried upgrading 18.04 to 20.04 but can’t get the Python 3 bindings for TensorRT due to that.
Probably going to use dusty’s info for building Noetic from source on 18.04 for now.
So I spent $700 on a computer and now I’m supposed to take another computer out of service so I can install an obsolete operating system on it just to install an SDK because you all can’t figure out how to build a self-booting computer the way Raspberry Pi people have been doing since forever?
Hi @jeff_science, the Jetson Xavier NX and Jetson Nano developer kits use microSD cards, which can be flashed from a PC running Windows/Mac/Linux, so you shouldn’t need to install another OS to your PC just to flash the SD card with the Etcher tool.
For reference, here is the guide to flashing SD card for the Nano:
https://developer.nvidia.com/embedded/learn/get-started-jetson-nano-devkit#write
The previous Jetson TX1/TX2/AGX Xavier developer kits use eMMC storage, which needs to be flashed over micro-USB from an Ubuntu PC using the SDK Manager tool. After flashing has been done once, you can upgrade to future JetPack versions using apt, without needing to reflash.
Yeah, I guess I’ll buy an SD and load the new image that way. However, what I learned from a friend is that I don’t need to do any of this. Jetpack is unnecessary to use the thing. The docs should really be improved to say “just boot it and start using CUDA immediately” rather than going into detail (there are lot of pointless tangents in the docs) about things that aren’t actually required.
That’s correct, if you are using the devkits for Xavier NX or Nano, then the SD card images come with all of the JetPack components pre-installed (including CUDA, cuDNN, TensorRT, ect). You only need to use SDK Manager if you are flashing one of the other Jetson’s, or if you are flashing a production NX/Nano module that use eMMC instead of SD card like the NX/Nano devkits do.
Is there any update on this timeline?
Hi @NicoMandel, no update about it yet, sorry about that.