NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation

[tokenwise]

Area of Concern

• [X ] Server
• Installer
• Server Dashboard or Explorer
• Installation issue of one or more Modules. Please post the issue on the module's Issue list directly
• Behaviour of one or more Modules Please post the issue on the module's Issue list directly
• Something else

Server Info

• CodeProject.AI 2.9.5
• Windows 11
• 16Gb
• NVIDIA GeForce 5060 Ti 16GB
• Using Docker? no

Describe the bug
NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 compute_37.

Expected behavior
It should work. Modern 50-series NVIDIA GeForce GPUs should be supported, just need to bump up to a more recent PyTorch to fix the issue

Additional context
An end-user cannot easily replace the PyTorch functionality in CodeProject

[tokenwise]

Please move this to the YOLOv8 module list

[bradpedersen]

Also fails with the same error using a Nvidia GeForce RTX 5070 with the "codeproject/ai-server:cuda12_2-2.9.7" docker image.

[uuzinger]

Same:

18:20:59:Preparing to install module 'ObjectDetectionYOLOv8'
18:20:59:Downloading module 'ObjectDetectionYOLOv8'
18:21:00:Installing module 'ObjectDetectionYOLOv8'
18:21:00:ObjectDetectionYOLOv8: Installing CodeProject.AI Analysis Module
18:21:00:ObjectDetectionYOLOv8: ======================================================================
18:21:00:ObjectDetectionYOLOv8: CodeProject.AI Installer
18:21:00:ObjectDetectionYOLOv8: ======================================================================
18:21:00:ObjectDetectionYOLOv8: Invalid number. Numbers are limited to 32-bits of precision.
18:21:00:ObjectDetectionYOLOv8: Invalid number. Numbers are limited to 32-bits of precision.
18:21:00:ObjectDetectionYOLOv8: Invalid number. Numbers are limited to 32-bits of precision.
18:21:00:ObjectDetectionYOLOv8: .Gb of Gb available on (No label) (Windows 11 x86_64 - windows)
18:21:00:ObjectDetectionYOLOv8: General CodeProject.AI setup
18:21:00:ObjectDetectionYOLOv8: Checking for VC++ Redist...v14 Present.
18:21:11:ObjectDetectionYOLOv8: Creating Directories...done
18:21:11:ObjectDetectionYOLOv8: GPU support
18:21:12:ObjectDetectionYOLOv8: CUDA Present...Yes (CUDA 13.0, No cuDNN found, CUDA Toolkit: false)
18:21:12:ObjectDetectionYOLOv8: ROCm Present...No
18:21:12:ObjectDetectionYOLOv8: Checking for .NET 9.0...Checking runtimes...All good. Found .NET aspnetcore 9
18:21:13:ObjectDetectionYOLOv8: Reading ObjectDetectionYOLOv8 settings.......done
18:21:13:ObjectDetectionYOLOv8: Installing module Object Detection (YOLOv8) 1.6.2 (Internal)
18:21:13:ObjectDetectionYOLOv8: Installing Python 3.9
18:21:17:ObjectDetectionYOLOv8: Downloading Python 3.9 interpreter...Expanding...done.
18:21:25:ObjectDetectionYOLOv8: Creating Virtual Environment (Local)...done
18:21:25:ObjectDetectionYOLOv8: Confirming we have Python 3.9 in our virtual environment...present
18:21:25:ObjectDetectionYOLOv8: Installing Python packages for Object Detection (YOLOv8)
18:21:25:ObjectDetectionYOLOv8: �[0;Installing GPU-enabled libraries: If available
18:21:26:ObjectDetectionYOLOv8: Ensuring Python package manager (pip) is installed...done
18:21:29:Response timeout. Try increasing the timeout value
18:21:32:ObjectDetectionYOLOv8: Ensuring Python package manager (pip) is up to date...done
18:21:32:ObjectDetectionYOLOv8: Python packages specified by requirements.windows.cuda.txt
18:21:37:ObjectDetectionYOLOv8: - Installing NumPy, a package for scientific computing...(✅ checked) done
18:22:11:ObjectDetectionYOLOv8: - Installing PyTorch, an open source machine learning framework...(✅ checked) done
18:22:54:ObjectDetectionYOLOv8: - Installing TorchVision, for working with computer vision models...(✅ checked) done
18:24:07:ObjectDetectionYOLOv8: - Installing Ultralytics package for object detection in images...(✅ checked) done
18:24:13:ObjectDetectionYOLOv8: - Installing the CodeProject.AI SDK...(✅ checked) done
18:24:13:ObjectDetectionYOLOv8: Scanning modulesettings for downloadable models...Processing model list
18:24:33:ObjectDetectionYOLOv8: Downloading YOLOv8 Object Detection all sizes...Expanding...done.
18:24:33:ObjectDetectionYOLOv8: Copying contents of objectdetection-coco-yolov8-pt-nsmlx.zip to assets...done
18:24:51:ObjectDetectionYOLOv8: Downloading YOLOv8 Object Segmentation all sizes...Expanding...done.
18:24:51:ObjectDetectionYOLOv8: Copying contents of objectsegmentation-coco-yolov8-pt-nsmlx.zip to assets...done
18:24:54:ObjectDetectionYOLOv8: Downloading YOLOv8 Object Custom models...Expanding...done.
18:24:54:ObjectDetectionYOLOv8: Copying contents of objectdetection-custom-yolov8-pt-m.zip to custom-models...done
18:25:54:ObjectDetectionYOLOv8: C:\Program Files\CodeProject\AI\modules\ObjectDetectionYOLOv8\bin\windows\python39\venv\lib\site-packages\torch\cuda_init_.py:155: UserWarning:
18:25:54:ObjectDetectionYOLOv8: NVIDIA GeForce RTX 5060 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
18:25:54:ObjectDetectionYOLOv8: The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 compute_37.
18:25:54:ObjectDetectionYOLOv8: If you want to use the NVIDIA GeForce RTX 5060 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
18:25:54:ObjectDetectionYOLOv8: warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))

[MikeLud]

Use the Object Detection (YOLOv5 .NET) module with the RTX5000 GPUs it will be faster then YOLOv8 and YOLOv5 also it does not require CUDA to be installed

[ChrisMaunder]

Time to update to YOLO12 I guess...

[MikeLud]

I have been working on on a YOLO11 License Plate Recognition module that uses onnxruntime-directml and ONNX models so we can move away from PaddePaddle

[ChrisMaunder]

As awesome is paddlepaddle is, it's only awesome when and where you can get it to work. It's fully in the too-hard basket for this sort of exercise so a change like this is very welcome!

[pilonull]

welllp, wish i would have known this before i upgraded to a 5060 ti... now it doesn't work. And the suggestion to switch to yolov5 .net module isn't great because it's CPU only and doesn't use GPU...

Seems codeproject has abandoned this project and no longer getting updates.

I found in another post it was possible to add additional modules but of course the link is now dead (https://www.codeproject.com/ai/docs/devguide/module_examples/adding_new_modules.html)

Great....

edit: found the updated link https://codeproject.github.io/codeproject.ai/devguide/module_examples/adding_new_modules.html but that doesn't seem to apply to docker installs...

edit2: updating this post in case other people have the same issue. MikeLud graciously built a new image available at https://hub.docker.com/r/mikelud/ai-server/tags that works great in docker on linux with 50 series cards and now uses yolov11.

Can read more about it here https://ipcamtalk.com/threads/new-codeproject-ai-object-detection-yolo11-net-module.83611/post-957933