oneAPI Registration, Download, Licensing and Installation
Support for Getting Started questions related to download, Installation and licensing for Intel oneAPI Toolkits and software development tools.

Disappearance of intel conda channel

Jacob_Williams
New Contributor II
2,890 Views

It seems that the intel conda channel is no longer available?

 

I.e., https://anaconda.org/intel

 

We used to be able to install the intel compiler runtime libs by using the intel channel in a conda environment.yml file like so:

 

channels:
  - conda-forge
  - intel
 
dependencies:
  - intel::fortran_rt=2023.0.0

 

But it's no longer working. Is this channel gone for good? Was there any announcement about that or what the recommended way forward is?

9 Replies
SamTygier
Novice
2,864 Views
0 Kudos
Jacob_Williams
New Contributor II
2,748 Views

 

 

For future reference: I originally posted this in the Fortran forum: https://community.intel.com/t5/Intel-Fortran-Compiler/bd-p/fortran-compiler

0 Kudos
Jacob_Williams
New Contributor II
2,747 Views

Inferring from this link: 

 

[then click installing from package managers, then conda]

 

It appears that Intel is now self-hosting their conda packages at: 

 

 
So, the anaconda/intel channel is no more? It would be nice to get some official word from intel about this. This has broken many workflows seemingly without warning. 

 

 

Trackah123
Beginner
2,411 Views

Had the same issue, thanks to your repo advice i managed to get it to work like this.

Example packagename "onednn"

conda install conda-forge::onednn -c https://software.repos.intel.com/python/conda/ --override-channels

0 Kudos
The_Hedgehog
Beginner
926 Views

Thanks, that's helpful!

That almost did it for me, but I also needed the following to get rid of issues while using Anaconda Navigator:

conda config --set allow_non_channel_urls True

Found that solution here:

https://stackoverflow.com/questions/78761316/conda-is-using-a-channel-after-i-removed-it 

It's not a 100% clean solution because this seems to tell the system to simply ignore the channel that isn't working but I tried other solutions and couldn't get rid of the issue otherwise. Even after removing the (old) intel channel, the system was still trying to use it for some reason.

0 Kudos
DManowitz
New Contributor I
900 Views

It seems like the issue you are encountering is the same as this one that I reported.  Is is an issue with the libmamba solver for conda.  You can set the environment variable CONDA_LIBMAMBA_SOLVER_NO_CHANNELS_FROM_INSTALLED=1, at least with conda-libmamba-solver>=23.9.0, and that should fix your issue.  You can also use the classic conda solver or use mamba itself, and you shouldn't see these errors.

0 Kudos
The_Hedgehog
Beginner
839 Views

Thanks for the excellent suggestion!

I tested once again, and first reverted the "set allow_non_channel_urls" change that I mentioned above. After that I set the environment variable CONDA_LIBMAMBA_SOLVER_NO_CHANNELS_FROM_INSTALLED=1, as suggested.

I confirm this indeed solves the issue with Anaconda, it then no longer tries to connect to the (old / non-existing) Intel channel, and can install packages again in a normal fashion.

0 Kudos
DManowitz
New Contributor I
1,919 Views

It would be nice to get some more official word of exactly what (and, ideally, why) this happened.  I also notice that the new repo doesn't have many of the older versions of various packages (or any Python 3.8 packages).

0 Kudos
Jacob_Williams
New Contributor II
1,052 Views

Oh Wow. I just noticed this too that some of the packages in the old channel are not in the new one.

 

Intel this is just totally unacceptable. Removing the old packages breaks deployment of anything that requires them. If we can't trust this channel to keep the packages there, we can't use this compiler.

 

What is the policy for retaining old packages? Are we in a situation where we can't rely on any package to stay there forever? 

Reply