From a user point of view (I mean me working on the machine as a dev), this is hopelessly outdated. Can I upgrade the ubuntu system in the normal ubuntu/debian way without breaking the nvidia specific and development packages? I also tried to look up later versions of the “jetpack” i.e. the system image, but the documentation pages for them seem to say they don’t support the nano at all (only Xavier and Orin).
Is this really so? In any case, I’d be happy with the original SDK for now, but I really need a more recent ubuntu.
This is probably the case. The SoC of the Nano is quite old, and new features are no longer developed (although it does get security fixes). It is possible that someone might know how to do this, but due to this (and most embedded systems) not having a BIOS, it means the boot code is custom, and most software out in the wild has no concept of Jetson boot code. In theory someone could adapt it, but the effort and knowledge required would be high.
Nvidia jetson nano is nothing old…
At the legal level of complaint, it was released in 2019 and has barely had two years of full support.
The GPU is based on Maxwell when in GPU PC Drivers it keeps getting implementations in the 510 branch when the nano is based on the very old nvidia drivers branch 440 or so? with wayland buggy support and old vulkan implementation.
We needed the same support as PC no mediocre support…
No Vulkan 1.3 (very import for near future), missing Wayland vulkan Working , A lot of bugs in resize or alt+tab hangs cause by nvidia egl core with no idea why.
Nvidia jetson nano could run Ubuntu 20.04 as a base without any problem excepted snapd packages gpu aceleration not detected correctly for some reason snap package run mesa driver first and not detected tegra libs.
Also needeed rebuild kernel with some flag for working
Anyway snap packages it’s optional option by canonical .
This support is very important for some applications.
This problem does not occur with Xavier or Orin blobs they can load vulkan wayland without fail.
Why not for Jetson nano? :/
This is my opinion for developing and expert linux system user
We also needed to install blobs on other distros due is Ubuntu repository being bug/old libraries for a while without testing
Debian repo it’s better than Ubuntu repos in perfomance and stability for SBC boards but the actual deb can install on debian due the compresion .
What will happen after 2023 EOL ubuntu 18.04? We will have a buggy paperweight base that we won’t be able to resolve without nvidia colaboration.
It isn’t the module format I’m speaking of. It is the SoC inside of it. This is the same SoC as used with the TX1, which in turn is much older. In fact, the TX1 was the first 64-bit ARM system with a GPU, and came right after the TK1 (which was in turn the first 32-bit ARM system with a GPU). The TK1 did not come with a separate module and carrier board; the TX1 was the first separate Jetson module/carrier board. The Nano was a new form factor, but the technology on the silicon was already old.
When you speak of the driver branch, such as the 440, you are speaking of PCI/PCIe-based GPUs. Those have different drivers than what Jetsons use due to the Jetson GPUs being directly integrated to the memory controller. The basic architecture, what you mentioned as Maxwell, is the same, but the drivers are incompatible in mainline versus Jetson due to the memory controller integration.
I think the GPU part of the drivers, excluding the memory controller versus PCIe, are probably the same. However, These older GPUs have not had new feature development for a long time, and during the time in which they were actively developed I think Wayland and Vulkan were much more primitive than they are now.
I know it isn’t what you want to hear, but if you want a more modern driver you would probably have to use one of the Xavier or Orin series (these are actively developed for new features and would tend to follow updates for Wayland and Vulkan). Even if there is a bug in software for Wayland/Vulkan in Orin or Xavier, this would be actively supported as a fix, whereas unless it is for security reasons, then it is unlikely to see such a fix with the maintenance-only SoC of the TX1/Nano.
This does not mean that there is not a bug fix possible for Wayland/Vulkan, but someone from NVIDIA would have to know of a specific error or issue to say if that has a fix. Upgrading to Ubuntu 20.04 won’t happen, at least not officially. I have seen people on the forums who sometimes succeed (with some work) at such an upgrade. Unfortunately, that upgrade won’t be an upgrade of the GPU driver since they are binary-only in these older systems, and going to a new Linux release might imply needing to either hold back the GPU driver release, or else going to Mesa (which is software-rendering, and loses the GPU’s benefits).
Installing kernel modules for various squashfs items should be relatively straight-forward. It is just a case of configuring that release of kernel source to your running system’s configuration (including CONFIG_LOCALVERSION), adding those features with a menu editor in module format, compiling, and copying the resulting squashfs “*.ko” files to the right subdirectory within “/lib/modules/$(uname -r)/kernel”. In fact there is an extended thread on adding this: https://forums.developer.nvidia.com/t/how-do-i-install-ubuntu-frame-in-jetson-nano/228370
(much of this though is just about how to compile and install modules; I don’t use squashfs, so I don’t know all of the requirements to be compiled when working on this)
The topic of supporting Xorg is in fact one of the more interesting topics. When it comes to a binary-only GPU driver in user space, then that means the driver has to work with the specific Application Binary Interface (the “ABI”), and that newer Xorg X servers https://forums.developer.nvidia.com/t/how-do-i-install-ubuntu-frame-in-jetson-nano/228370usually use a new ABI. If a newer Xorg server were to be available with the older ABI, then there would be no issue to simply copying the old GPU driver from the older release into a newer release.
There was some semi-recent news about NVIDIA providing source now for GPU drivers, but I don’t know if that applies to Jetsons. I don’t know the details, but at least for the PCIe video cards, there is apparently some publicly available source for it now.
I know how it works and the differences that may exist
But that’s NVIDIA’s problem only.
It’s no excuse not to update it
What I’m saying is that Maxwell PCI express GPUs can support things at another level of support by drivers that nvidia jetson TX1/TX2 can’t and abandoned why not reason
A hardware has to support everything that these hw supports regardless of first architecture how old it is
If they don’t want to do it, then let them release the drivers and someone else will…
Well collaborate early has created NVK driver for mesa3d from zero Introducing NVK
But will be slow and limited support for before Ampere if i’m not wrong and early 1.0/1.1 partially stuff for now…
Because I could speak clearly at first Nvidia jetson boards are products with very poor support in the medium and long term.
And here they are objective tests, nothing subjective.
And the host tools do not make sense to tie Ubuntu knowing how problematic Canonical is in Linux support, it is one of the most hated companies and very poor repos.
If you tell me you prefer the apt system Well, run Debian base image the community is much better and you have a white sid for development and daily use but Please no base Ubuntu support is very pain for current standard development and fuck snap or latest compilers/libs.
And worst of all isn’t that people know what they will do? Going to Qualcomm with Adreno 650 released in 2019 and now supporting Vulkan 1.3 and mainline support.
Mostly have can support vulkan 1.2/1.3 for broadcom rpi4 very weak gpu too than TX1/TX2 and mainline support.
If Qualcomm can implemented for old chip in 2019 released and moslty for rpi4 too why not Nvidia too?
If I were responsible for the hw, I would be embarrassed that less powerful gpus support some things that this hw does not intend to implement i.e gpu rpi4 supported some extensions than TX1/TX2 can’t not due very old implementation.
If the other companies do it, nvidia to take an example and do it too.
What will people do? the people will forgive 1 or 2 times but the third time he will go to the competition that offers the best support for latest development same as PC use.
I have always defended nvidia and on pc whenever I can I use nvidia for the extra support and multithread performance in OpenGL and Vulkan when things are working specially in emulators and gaming.
But when it stops supporting that HW it becomes a paperweight in view of future support.
Nvidia must change its mind if it doesn’t want to lose customers who use its boards for daily development and a daily SBC for all things
Ah DXVK Requires now Vulkan 1.3 so will never work under TX1/TX2 in the actual/future stuff :-) Thanks NVIDIA :D