search

ContinuumIO / anaconda-issues

Anaconda issue tracking
646 stars 220 forks source link

Conda on Nvidia Jetson tx1 tx2? #1587

Open roryw10 opened 7 years ago

roryw10 commented 7 years ago

A bit of a long shot here but i'm wondering if anyone has experience getting Anaconda working on an Nvidia Jetson tx1 or tx2? I understand it's a bit of an unusual requirement. The distro is linux for tegra which looks to be ubuntu 14.04 set up with some preconfigured drivers...

"Linux version 4.4.15-tegra (buildbrain@mobile-u64-1423) (gcc version 4.8.5 (GCC) ) #1 SMP"

I thought that given its linux 64 running ubuntu that conda should run however at the end of running the bash script i get a lovely error

'Anaconda2-2.5.0-Linux-x86_64.sh: line 453: /home/ubuntu/anaconda2/pkgs/python-2.7.11-0/bin/python: cannot execute binary file: Exec format error ERROR: cannot execute native linux-64 binary, output from 'uname -a' is: Linux tegra-ubuntu 4.4.15-tegra #1 SMP PREEMPT Wed Mar 1 21:09:29 PST 2017 aarch64 aarch64 aarch64 GNU/Linux'

I understand this is not a common device / set up and i can probably get by without using Conda however it definitely makes my workflow much easier.

Thanks

mingwandroid commented 7 years ago

I'd love to tackle this, but it would be a non-trivial amount of work so we'd probably need a customer to pay for us to do it.

jjhelmus commented 7 years ago

I believe both the TX1 and TX2 both use ARM CPUs and therefore the the standard x86 and x86_64 Anaconda will not work.

It may be possible to install the armv7l Miniconda, although I have my doubts as it targets the Raspberry Pi 2/3s which use ARMv7 instruction set rather than the ARMv8 set that the Tegra's use. If it does work please know that the packages in the repository for that platform are quite old and are not being updated.

mingwandroid commented 7 years ago

AArch64 can run armv7a just fine, but you aren't taking full advantage of the CPU (far from it). A proper Tegra port should be AAch64 only, use neon for floating point and should try to take advantage of Cuda where possible.

roryw10 commented 7 years ago

Sorry, have not been able to look at this till now. @jjhelmus slight issue in that the current download link for linux 64 python 2 is hitting a 404....

https://www.continuum.io/content/conda-support-raspberry-pi-2-and-power8-le

https://repo.continuum.io/miniconda/Miniconda-latest-Linux-ppc64le.sh

Same for python 3 (though would prefer to stick with 2 for now) - 32 bit versions appear to be working!

jjhelmus commented 7 years ago

@roryw10 Those broken links are to Miniconda installers for the ppc64le architecture. They would not work on the NVidia Jetson.

elehcim commented 7 years ago

With conda/conda#5190 implemented, is it possible to try to have an installer for aarch64 like the Jetson?

thomasantony commented 6 years ago

@elehcim https://github.com/conda/conda/pull/5190#issuecomment-319774238

yihuajack commented 2 years ago

It seems that the latest install script Anaconda3-2021.05-Linux-aarch64.sh does not support Jetson TX2. After installation, any conda commands like conda info conda list would fail with error: [1]    11034 illegal hardware instruction (core dumped)  conda info