The problem is very,very frustrating and I think basic.
Many of us is using a lot of different linux version. Each of us has favorite linux OS and can install on them necessary programs.
Jetson Nano is bought NOT only for AI and Neural Network Learning…is bought also as a SBC desktop for dayly using…and many of linux users use sd/usb booting…as only a hardware donor
and USB pendrive is only dayly desktop software carrier for normal working.
If Jetson Nano want to be as good and popular as Raspi or Atomic Pi…or other SBC x86…must be in some things better. This is the basis of competition with raspberry pi and x86.
Rasp Pi have ARM-hf processor. Hf means floating-point processor calculation instructions.
Non Hf processors like Inte/Amd x32 or x64 has fixed poind processor calculation instructions.
I have readed that some of Arm=hf processors and some newer Linux Intel/amd x32/x64 distros are fabiric adapted to hf. I mean that normal iso image what we downlaoded from linux os sites can be mount on pendrive USB and boot normaly like on standard x64 if its x64 or on x32 if we download x32 version.
I am using Echer or similar distro based on ISOlinux…to write iso image on usb pendrive and boot them from usb. On Rasp pi…there is converter barry boot what allow to convert “normal” linux iso to arm-hf iso version. But it is very frustrating, time-consuming and non practical because…most of linux usb boot users…want to and use…only one pendrive witch favorite linux to dayly using desktop…for example x64… put…to many …many…computers…what they have…or where they exactly are… I know that Ubuntu…prepared by Nvidia has many dedicated AI and DL, Neural programs…but…normal users want to also have a normal desktop ability to run on Jetson each…linux distro.
What Nvidia will do to solve the problem?? and makes live Jetson users live batter??
Mmaybe nvidia developers contact to barry boot owner to make some…script what will be automaticly…run…any x64 (amd/intel) distro linux…I mean …calculated on the fly non hf instructions to hf instructions…in isolinux low-level code overlay.
I am just wondering. Maybe some low layer hf to no hf in fly live emulator?
I dont know I am not a programist.
I want to …run my x64 favorite linux distro what I have on USB pendrive…and I always booting on many computers…when I will put pendrive to usb…
This function have for example SBC Atomic Pi and some others x86/x64 SBC.
I dont want to use addidtional Linux converted HF version.
I want to have only one usb pendrive witch one favorite normal standard Linux…booting from SBC.
Dear Nvidia Stuff…what do you do…to makes a dream to made a open single board computer to be the home of any iso distro linux…run…free…to fell happy:). If you will do this…Nvidia Jetson Will be batter than Rasp Pi 4… and many of OpenSoftware and foss:) fans…will be happy:)
The competition is not sleeping
- Atomic Pi
all of this has able to run “normal” x64 and even! x84 Linux version without any freakse.
Do you look on this matter? and take it seriously or not?
I want to know.
I don’t believe there is a specific answer possible, but here are some semi-random concepts which will probably be useful to you.
FYI, for reference, armhf is 32-bit with optional hardware floating point, which is sometimes named just “arm” or “arm32” (ARMv7). 64-bit ARM is one of “arm64/aarch64/ARMv8-a”. All ARMv8-a has a 32-bit compatibility mode, “ARMv8”, but that compatibility mode is not normally run and absolutely sucks for performance. Additionally, the compatibility mode always comes with hardware floating point, and thus no compatibility option…hf is always there.
All x86 is incompatible with anything ARM, with no possibility of the two being interchangeable. Most ARMv8 32-bit compatibility will run on ARMv7, but not all of it; all ARMv7 will run on ARMv8 compatibility mode…but only if you’ve installed an enormous amount of user space support and can live with terrible performance.
Neither RPi nor Jetson will ever run an x86 package without a VM environment (and absolutely terrible performance).
The part which sets embedded systems apart from a desktop PC is the lack of a BIOS/UEFI. This interface does a lot for setting up clocks and power regulators, so on, and is why Windows and Linux do not need a custom version for every single motherboard made. Embedded systems must do this setup in software, and thus use custom software setup for the boot stages prior to starting the operating system; every single carrier board or SBC made must have this customized and is not portable across SBCs. Most of the operating system itself is portable among same architecture, but firmware is likely needed to tell the system where to find different parts of the hardware.
The part where a Nano exceeds the RPi by a lot is with the GPU. The software surrounding the GPU is custom to NVIDIA, and without this CUDA and other hardware-accelerated functions cannot occur. Some people have ported the various arm64 Linux releases/distributions to Jetsons, and it works, but rarely will the CUDA or GPU-related functions work. In those cases where it did work the Xorg X11 server was kept at the specific ABI release required for the NVIDIA GPU driver to be loaded by the X11 server, but even for people who can port entire operating systems it is not trivial to be able to make sure the X11 server supports a specific ABI.
It won’t be possible to do what you want, either on Pi or Nano. X86 (pc) binaries are not compatible with arm, and even within arm and x86 there are compatibility issues.
If you want to run linux from a pen drive that boots on every single architecture out there, such a thing does not currently exist and won’t likely ever be possible.
Atomic Pi will work because Atomic Pi is x86 (Intel Atom). Same with Udo and LattePanda. There are hardware virtualization extensions on some x86 processors that will allow you to run arm code very quickly, however, the same is generally not true the other way around.
FWIW, I don’t think normal desktop usage is the target market for the Nano. It’s intended for things like kiosks, headless surveillance monitoring, anything IOT that needs to do GPU work at the edge (the other boards would probably perform poorly in comparison and don’t have the same level of support). The CPU is not intended to be very powerful… Just powerful enough for most needs, and it is.