Intel ARC / IPEX Docker support

[kmscode]

Description

• Print out the docker boolean properly
• "Add" support for IPEX supported python versions
• Support downloading IPEX from the US server via "SD_IPEX_USE_US_SERVER" variable
  • downloading from the CN server failed for me
• new Dockerfile.arc to create IPEX supporting docker image

Notes

IPEX supports 3.9-3.12:
https://pytorch-extension.intel.com/installation?platform=gpu&version=v2.5.10%2Bxpu&os=linux%2Fwsl2&package=pip

Environment and Testing

• Docker image base of Ubuntu 24.04 running on Unraid 7.0 (6.6.68-Unraid)
• Intel ARC A310 - which appears to have the same feature support as higher end ARC GPUS (A7xx and A5xx): https://www.intel.com/content/www/us/en/docs/dpcpp-cpp-compiler/developer-guide-reference/2025-0/ahead-of-time-compilation.html#id-d244160e597

This was my first attempt into playing around with genai/image generation - not too familiar with all of the sampling methods, but tried a few and some differing results as can be seen in the screenshots below.

Some images images showing process/processing and gpu usage and one output file:

[vladmandic]

thanks for the pr, few notes below...

i'd be ok with:

• expanding allowed python versions,
• fix for docker status
• switching default download location to us and use of env variable to use cn instead if needed

the rest?

• pr should never include submodules: right now you have both
  extensions-builtin/sdnext-modernui and extensions-builtin/sdnext-modernui which MUST be excluded
• pr should always target dev branch, never master
• inclusion of Dockerfile.arc in the main repo is not something i want to have
  we can have it in wiki or something?

cc @Disty0

[kmscode]

@vladmandic Thanks for the feedback!

I'll fix the submodules. It seems they got pulled in when I did the latest merge from main best I Can tell.

The CONTRIBUTING file was a bit unclear - it said to fork from main, so I assumed pr should also go to main.

I was planning on updating the wiki as well, so I'll drop the Dockerfile.arc there.

Expect an update.

[vladmandic]

@vladmandic Thanks for the feedback!

I'll fix the submodules. It seems they got pulled in when I did the latest merge from main best I Can tell.

The CONTRIBUTING file was a bit unclear - it said to fork from main, so I assumed pr should also go to main.

I was planning on updating the wiki as well, so I'll drop the Dockerfile.arc there.

Expect an update.

ah, contribution file needs an update - thanks for pointing in out.

[kmscode]

@vladmandic this should be good for further review now.

I added a quick note to CONTRIBUTING as well.

I had some edits/updates for the wiki in progress, I'll get those completed and include the updated ARC specific docker file there as well. That may be a few more days.

cc @Disty0

[Disty0]

switching default download location to us and use of env variable to use cn instead if needed

Only reason we are using the CN location is because US doesn't work right now. US server is broken on Intel's side and throws 403 errors when trying to download ipex 2.5.

We can simply switch back to US if this issue is fixed on Intel's side.

[Disty0]

Also i do have my dockerfile that only has the bare minimum for ipex if you are interested: https://github.com/Disty0/docker-sdnext-ipex/blob/main/Dockerfile

[vladmandic]

US server seems to be working now - can you double-check?
if that's the case, i'd rather have US as default instead of CN.

[Disty0]

US server seems to be working now - can you double-check? if that's the case, i'd rather have US as default instead of CN.

I agree with US being the default, download speeds of CN server is terrible.

Seems like Intel updated their installation guide to use US server again but main issue on the ipex repo is still open: intel/intel-extension-for-pytorch#745

US server seems to be working fine on my end too, we can set US as the default or just switch back to US instead of adding a new env variable.
Users can use TORCH_COMMAND if they want to change it.

[vladmandic]

US server seems to be working fine on my end too, we can set US as the default.

lets do that. no need for CN-specific env variable as users can use TORCH_COMMAND env variable already

also, lets not edit CONTRIBUTING in this PR, its unrelated plus i want to make larger changes there.

[vladmandic]

@kmscode thanks for updates
@Disty0 pls take a look and if you're ok with this, merge

[kmscode]

@vladmandic got this PR down to the bare minimum per the conversion notes above. Should be ready for review again.

[Disty0]

Please also revert the Python version changes. It shouldn't log 3.9-3.12. Seeing 3.12 instead of 3.9, 3.10, 3.11 will lead people to install 3.12 and that will cause more compatibility issues.

[vladmandic]

true. python 3.12.3 works with cuda, but anything newer than that breaks pydantic and tons of other totally unrelated packages.