Mobile and Desktop Processors
Intel® Core™ processors, Intel Atom® processors, tools, and utilities
Announcements
For support on Altera products please visit the Altera Community Forums.
17261 Discussions

Seeking Technical Assistance: Intel Ultra 9 NPU Integration with PyTorch on Windows 11

YUANZHU
Beginner
4,632 Views
I am writing to request assistance with a persistent and blocking issue regarding the use of the Intel Ultra 9 NPU (AI Boost) for accelerating PyTorch model training on a Windows 11 system. My core, non-negotiable requirements are: Primary OS must be Windows (not WSL or Linux). Models must be developed and trained using native PyTorch. The installed Intel AI Boost NPU hardware must be utilized. The problem I face is that while the NPU is correctly recognized by Windows Device Manager and the task manager, I cannot make it work through the expected software paths: intel-extension-for-pytorch is not an option as it explicitly does not support Windows. OpenVINO Toolkit encounters a critical failure. When attempting to compile and run a Transformer model (exported from PyTorch via ONNX), the process crashes with errors pointing to core.cpp and plugin.cpp, indicating a likely low-level driver or runtime incompatibility. The system may become unstable or the process terminates unexpectedly. My attempt to enable the NPU within WSL2 also failed, as the /dev/intel_npu* device node was not present, suggesting the driver does not correctly pass through. I have ensured the latest Intel NPU driver is installed. The core contradiction I am stuck with is the lack of a supported, stable software bridge between PyTorch on Windows and the NPU hardware. Could you please advise: Is there a confirmed, stable software stack (specific driver + framework versions) for this use case? Are there any known workarounds or configuration changes for the OpenVINO runtime crashes on Windows 11? Is the newly open-sourced Intel NPU Acceleration Library intended to address this exact gap, and what is its current readiness for production PyTorch workloads? Any guidance or insight into resolving this "Windows + PyTorch + NPU" trilemma would be greatly appreciated. Thank you for your time and support. Sincerely, A Developer
0 Kudos
7 Replies
JohnM_Intel
Moderator
4,592 Views

Hello YUANZHU,


Thank you for raising this concern via the Intel Community.


To properly investigate the issue, I would like to gather some information first so I can provide you with the appropriate steps to help resolve it. Kindly provide the details below:

  • Are you using a laptop or a desktop system?
  • What is the exact model of your Intel® Core™ Ultra 9 processor?
  • Does the crashing occur only when using any OpenVINO™ applications?
  • Are all system drivers, BIOS, and software fully up to date?


Once the information is provided, it will greatly help with our investigation.


Thank you, and I look forward to your response.


Regards,


John M.

Intel Customer Support Technician


0 Kudos
JohnM_Intel
Moderator
4,535 Views

Hello YUANZHU,


I hope you had the opportunity to review the information I posted. At your earliest convenience, please let me know so we can determine the best course of action to resolve this matter efficiently.


Regards,


John M.

Intel Customer Support Technician



0 Kudos
YUANZHU
Beginner
4,532 Views

Dear John M.,

Thank you for your response.

Please find my answers below:

1. Are you using a laptop or a desktop system?
I am using a laptop.

2. What is the exact model of your Intel® Core™ Ultra 9 processor?
Intel(R) Core(TM) Ultra 9 185H (2.50 GHz)

3. Does the crashing occur only when using any OpenVINO™ applications?
Yes, the crash occurs specifically when using OpenVINO to compile a model for the NPU device.
The NPU is detected ("NPU" appears in ov.Core().available_devices), but calling core.compile_model(ov_model, "NPU") fails with errors referencing core.cpp and plugin.cpp.
OpenVINO operations on CPU work normally.

4. Are all system drivers, BIOS, and software fully up to date?
All related software and drivers have been updated to the latest stable versions available through official channels at the time of testing.


Please let me know if any additional information or logs are needed.

Thank you again for your assistance.

Best regards,
YUANZHU

0 Kudos
BrenzL_Intel
Moderator
4,496 Views

Hello YUANZHU,


Thank you for this information.


I will do further research on this matter and post the response on this thread once it is available.

 

If you have questions, please let us know. Thank you.


Best regards,

 

Brenz L.

Intel Customer Support Technician


0 Kudos
BrenzL_Intel
Moderator
4,426 Views

Hello YUANZHU,


Thank you for your continued cooperation with Intel.


Please install the correct NPU device driver by following the instructions provided in the NPU configuration guide.


After installing the proper driver, proceed to Install the distribution of OpenVINO™ toolkit via OpenVINO Archives.


Once the installation has been completed, please source the environment variables from the OpenVINO installation, also available through the OpenVINO Archives.


If you need further assistance with any of these steps, please let us know.


Best regards,

 

Brenz L.

Intel Customer Support Technician


0 Kudos
BrenzL_Intel
Moderator
4,326 Views

Hello YUANZHU,

 

Could you please confirm if you have reviewed the information I posted? Your feedback at your earliest convenience would be greatly appreciated so we can decide on the best way to proceed with resolving this matter.

 

Best regards,

 

Brenz L.

Intel Customer Support Technician


0 Kudos
BrenzL_Intel
Moderator
4,280 Views

Hello YUANZHU,


Since I haven't received a response from you, I will be closing this inquiry. If you need further assistance, please submit a new question, as this thread will no longer be monitored.


Best regards,

 

Brenz L.

Intel Customer Support Technician


0 Kudos
Reply