How to use AMD HX370's NPU to run yolo11 ?

[Lennox7746]

I have a laptop equipped with the HX370 chip, and AMD says it has 50 TOPs of NPU computing power. I want to use this NPU to run the YOLO11 model for real-time inference. What should I do? I couldn't find any relevant information online.

[UltralyticsAssistant]

👋 Hello @Lennox7746, thank you for your interest in Ultralytics and YOLO11 🚀! We recommend checking out the Docs for detailed guidance on using YOLO11, including examples for Python and CLI. These resources might help clarify some of your questions.

If this is a hardware compatibility ❓ question, please ensure that your system meets all requirements mentioned in the Installation Guide and verify that the necessary drivers and frameworks for your AMD NPU are installed and working correctly.

For specific hardware integrations like the AMD HX370 NPU, you may need to check if AMD provides compatible frameworks or libraries (e.g., ROCm or other tools) that support PyTorch or YOLO models. If you are encountering any issues, providing detailed logs or error messages can help us assist you better.

Join the Ultralytics Community

Engage with the Ultralytics community for additional support and insights. You can:

• Join the Discord server 🎧 for real-time discussions.
• Explore the Discourse platform for in-depth topics.
• Share ideas and tips on the Subreddit.

Upgrade Instructions

To ensure you're using the latest features and fixes, upgrade the ultralytics package in a Python>=3.8 environment with PyTorch>=1.8:

pip install -U ultralytics

Verified Environments

YOLO11 can be run in various verified environments. If you'd like to test it in a pre-configured setup, try one of the following:

• Notebooks:
• Cloud Services: Check the GCP Quickstart Guide or AWS Quickstart Guide.
• Docker Image: Refer to the Docker Quickstart Guide .

Status

If the CI badge is green, all Ultralytics CI tests are passing, ensuring compatibility across all YOLO Modes and Tasks on supported platforms.

This is an automated response, but an Ultralytics engineer will review your question and provide further assistance soon. Let us know if you have additional details or logs to share! 😊

[glenn-jocher]

Thanks for your interest in deploying YOLO11 on AMD hardware! While we don't currently have direct NPU integration for AMD platforms, you can explore converting YOLO11 to ONNX format (see our ONNX Export guide) and using AMD's Ryzen AI software stack for NPU optimization. For implementation specifics, we recommend consulting AMD's developer resources or community forums, as NPU utilization often requires platform-specific toolchains. Let us know if you have other questions about YOLO11 capabilities!

[Lennox7746]

OK，I am trying according to AMD's routine. Thank you very much for your reply. Thank you

[glenn-jocher]

@Lennox7746 you're welcome! For AMD NPU deployment, we recommend starting with the YOLO11 ONNX Export guide and then exploring AMD's Ryzen AI tools for hardware optimization. Feel free to reach out if you encounter any specific technical hurdles during implementation. 🚀

[kvrban]

Is there an update here? It would be very interesting to know whether the NPU in the new AMD chips can be used to accelerate YOLO.

Update:
just found:
https://github.com/amd/RyzenAI-SW/tree/main/tutorial/object_detection

do not have a corresponding AMD chip yet. But looks promising

UPDATE2:
Traffic Analysis using optimized YOLOv8 with AMD Ryzen AI:
https://www.hackster.io/jallsonsuryo/traffic-analysis-using-optimized-yolov8-with-amd-ryzen-ai-20fa29

[Lennox7746]

thanks very much

[Y-T-G]

You need to install this first

https://ryzenai.docs.amd.com/en/latest/inst.html

And then export your model to ONNX:
https://docs.ultralytics.com/modes/export/

And then run inference with the ONNX model but add this code:

import onnxruntime as ort

_InferenceSession = ort.InferenceSession
def InferenceSession(*args, providers, **kwargs):
    return _InferenceSession(*args, providers=["VitisAIExecutionProvider"], **kwargs)

ort.InferenceSession = InferenceSession

from ultralytics import YOLO

model = YOLO("yolo11n.onnx")
results = model()

[ESchouten]

@Y-T-G Ultralytics still seems to try and install onnxruntime-gpu, which seems to be the Nvidia driver

requirements: Ultralytics requirements ['onnx', 'onnxruntime-gpu'] not found, attempting AutoUpdate...

How did you overcome that?

[Y-T-G]

It would only install that if you pass device as GPU. Like device=0

[Y-T-G]

You should pass device="cpu"