Nvidia Jetson Nano

Offtopic
21

Gunter, the Linux/Ubuntu installation for the Jetson Nano is very up to data and is easy to use.
You can install Julia by using snap install julia - but as discussed elsewhere on here today that will give you Julia version 1.0.4

It is your all of course, but if you want to try Ububtu we can help and support you.

22

Hello John, thanks for your support! The page ā€œGetting Started With Jetson Nanoā€¦ā€ offers a tutorial for Windows and Iā€™m working on it now! At the moment Iā€™m trying to get the fan on the cooling fins and as always, the screws donā€™t fit! :grinning::sob:

I will report about the state of progress and if I have problems, I will gladly come back to your offer! Thanks a lot!

23

I believe the fan is not really needed until you start using it at maximum.
Also one tip - make sure to use a power supply which will work with a Raspberry Pi or similar.
A feeble mobile phone charger plus a cheap cable will not do it. You find this out when the Nano shuts down soon after booting.

I mention cheap cables - cheap cables are cheap because they use less copper, and so there is a voltage drop when high current goes through them. Select a decently thick cable if you can.

24

Thanks for the hint! I bought the Nano in a bundle (Nano, fan and power supply). It is offered in Germany by a reputable provider (Heise, very well-known magazine in Germany is the cā€™t) and still assumes that the power supply is the right one. This afternoon the first booting takes place! :champagne:

25

I like to admit that I am a Windows user. And since a few days, Iā€™m trying to get used to Ubuntu to use my Jetson Nano with Julia. Itā€™s a horror because Iā€™m used to the luxury of Windows! :sunglasses:

But seriously, did anyone get Julia version 1.0.4, which is offered through Ubuntu software, to work? I ā€œonlyā€ see a ā€œheartbeatā€ of the symbol for 15 seconds and then nothing happens anymore! Simply nothing more! Is Julia not compatible with the Jetson Nano? :roll_eyes:

26

Julia runs on the Jetson Nano! After Iā€™ve gone a little deeper into tar.gz file handling and struggled with the path variable, it works! Of course the current version and without compiling any files. The Nano was directly ā€œready for useā€! :grinning::champagne:

The exemplary package installation also worked directly!

Bildschirmfoto%20vom%202019-11-19%2018-41-08
Bildschirmfoto%20vom%202019-11-19%2018-41-08746Ɨ471 217 KB

Thank you for your time and support!

7 Likes
27

Has anyone tried playing around with ENV["CUARRAYS_MEMORY_LIMIT"] on the nano with CuArrays? Itā€™s rather memory limited, so Iā€™d be keen to see if anyoneā€™s found a sweet spot

28

Iā€™m not there yet! I still stumble across the package installation, e.g. Arpack, and am now thinking about a workaround. Next, I want to install Flux, hopefully it will work without any challengesā€¦

29

Couldnā€™t be more of a beginner honestly but I managed to get Julia working on the Jetson Nano and run the above code.
Result:
97.924 ms (18 allocations: 3.82 MiB)

4 Likes
30

Welcome to the community!!!

Did you run into any snags with the Jetson Nano?

31

Thanks a lot for make this tests! Are you generally satisfied with the performance for your application?

32

Thanks! Not really. I had a lot of command line learning to do in
Linux as a noob but once that was done it went pretty smooth. I
did have more luck downloading from Julia and going through the
steps to move the files to the proper folder and making the binary
executable.

1 Like
33

Today I finished a video (in German language) that shows the ā€œnewā€ Jetson Nano user how Julia is installed and used. I would be happy if especially Windows users find the video helpful.

5 Likes
34

thereā€™s a new model now:

2 Likes
35

I installed following this post. Installation of v. 1.5 went smooth in general; I just needed to install gfortran as a dependency.
Looking forward to run some code on it! :crazy_face: :nerd_face:

2 Likes
36

is anybody check julia in jetson nano 2GB kit ? is any memory problem ?
kindly suggest :slight_smile:

37

work well with jetson Nano 2GB :slight_smile:

Screenshot from 2020-11-28 14-53-00
Screenshot from 2020-11-28 14-53-001366Ɨ768 305 KB

4 Likes
38

My random thoughts:

  1. The training should be done by a host, with the Nanoā€™s delegated to recognition in the field.
  2. Be interesting if we could train the identical datasets on both a desktop and a Nano for a time comparison.
  3. IMO, itā€™s absolutely required that all flux be run in GPU mode as long as thereā€™s not a memory constraint. That is, the model actually fits onto the Nano. That severely limits our model sizes, comparing the 2/4 GB on the Nano versus 64 GB on a host.
  4. L-O-N-G compiles could occur on the by cross-compiling on the host. Host in this case being a strong PC desktop or workstation. Eclipse would probably work and maybe MVSC (Visual Source Code?)
  5. I have a 2 (coming) and a 4GB Nano as well as a TX2. When I win the big lottery, Iā€™ll get an Xavier for all these tests.
  6. The TX2 and the Xavier can use NVMe disk as virtual memory with the proper boot setup. Not sure if the Nanoā€™s will ever have that capability. :frowning:
39

Also, methinks the major reason to use a Nano versus an RPi is for that sweet, onboard GPU. Yeah, I know, shooting or the starsā€¦

40

You can ā€œfugdeā€ the memory on the Jetsons by installing a fast USB drive of considerable size, then getting a few gigabytes of it as virtual memory.