Kinara ARA-SDK Licensing

[NoNu]

Hi NXP team (and anyone familiar with Ara-2 / Kinara tooling),

I’m trying to understand the correct licensing/access path for compiling models for the Ara-2 (Kinara) compute module, especially now that Kinara has been acquired by NXP and NXP’s public materials describe SDK integration into NXP’s software ecosystem.

My setup / goal

• Hardware: Geniatech Ara-2 compute module
• SDK: Geniatech-provided Kinara Ara-2 SDK r1.3
• Use case: personal, non-commercial (research/study/testing)
• Model target: Qwen/Qwen2.5-7B-Instruct (GPTQ Int4)

What works so far

• I can complete the Mode 1 flow to convert the model into ONNX.
• I end up with:
  • model.onnx (small graph file)
  • model.onnx.data (large external tensor data / weights) in the same directory.

What fails

• Mode 2 (the compilation stage to generate a deployable .dvm) fails with license checkout/validation errors from the SDK’s build tools.
• From what I can tell, this blocks generating a .dvm even though ONNX export succeeds.

Where I’m stuck

• I do not have a Kinara license key.
• Geniatech’s documentation indicates that obtaining the SDK license key must be done via Kinara.
• The Kinara customer support portal appears to reject registration using common personal email domains (e.g., gmail.com), so I can’t submit a request there.

Why I’m asking here (NXP/Kinara integration confusion)
NXP’s public material indicates Kinara’s SDK and model tools are being integrated into NXP’s ecosystem (including eIQ), and Ara-2 is now presented as an NXP product offering with related “Ara Software Development Kit” downloads.
However, in practice, it’s not clear:

• where a developer is supposed to obtain a license key for compilation, and
• whether the “Ara SDK / eIQ integration” path is the supported route now.

Questions

1. Is a license key required to compile ONNX → .dvm using the Ara-2 SDK toolchain (r1.3 / ARA-SDK), even for personal/non-commercial use?
2. If yes, what is the official process for an individual/hobbyist user to obtain evaluation/developer access (especially without a corporate email domain)?
3. With an existing NXP account, where exactly should Ara-2 users obtain:
   • the “official” Ara SDK binaries/toolchain (compiler), and/or
   • the licensing mechanism needed for compilation?
4. If compiling requires a license key that isn’t readily obtainable, are there any precompiled .dvm model packages (e.g., for Qwen 7B/Qwen2.5 7B) that NXP/Kinara/partners provide for Ara-2 users to run?

If someone from NXP/Kinara can clarify the intended path (NXP portal vs legacy Kinara portal, and how licensing is handled now), that would help a lot.

[NoNu]

@kratafila mine was also basically unused silicon until I managed to get hold of a Linux SDK bundle via Geniatech. To be honest it is still unused silicon since I cannot get the models to compile and run.

For what it’s worth, Geniatech (vendor of my Kinara Ara-2 M.2 module) shared these SDK/runtime downloads with me. They’re Linux-focused, and I’m not sure whether they’ll apply to the Ara-2 hardware inside your ThinkCentre (or help on Windows 11, there are included Windows binaries but I had more success with the Linux software), but they might still be useful if you can test under Linux / confirm your device works:

1) Drivers/runtime + sample .dvm models (Geniatech "customer ready" bundle):

hxxps://mega[.]nz/file/nJcF0K5a#W-Ote-fp59hXoq4T0GGsaQmwGTRphWz0JATowyWjQpg

2) "Model compile" folder Geniatech initially sent me (note: this alone didn’t include the compiler binary):

hxxps://mega[.]nz/file/KoclFQrJ#ifNOX7w2Y1qgLM6rnm7xPnUprwZZqhuRvelFG5p0MJQ

3) Full Ara-2 SDK tarball Geniatech later provided (this is the one that should include the actual toolchain/compiler, e.g. dvrun):

hxxps://file[.]geniatech[.]com/down-eng/BSP/kinara_SDK_20251120[.]tar[.]bz2

If you do try any of the above and make progress (even just getting the SDK installed and seeing the device recognized), please post back here, not just for us, but because there seems to be broader confusion across multiple communities about the "official" working path and where the SDK/licensing is actually meant to come from as well as how to compile and run additional models.

[kratafila]

I have the same question! I got the device, driver. But no SDK and I guess no license either (if it is needed). The device is just sitting silently in my Lenovo ThinkCentre Ultra neo computer and not doing anything. Lenovo never finished its integration to Windows 11 as an NPU. I would like to use it at least to run some models on it or learning. It is now a completely unused piece of silicon in my PC. I guess it would help the adoption of the technology if more people would start using it.