YoloSharp.Gpu 6.1.0 not working on GTX 1080 Ti (GPU crash)

[Alex-code24]

After updating YoloSharp.Gpu from version 6.0.7 to 6.1.0, GPU execution on NVIDIA GeForce GTX 1080 Ti stopped working.

Version 6.0.7 works perfectly on my GPU. The problem appears only in 6.1.0.

Error Message:
onnxruntime::ExecuteKernel] Non-zero status code returned while running QuickGelu node.
Status Message: CUDA error cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the device

Steps to Reproduce:
Install YoloSharp.Gpu 6.1.0
Run any YOLO model on GTX 1080 Ti
Observe the CUDA kernel error

Expected behavior:
Models should run on GPU as they did in version 6.0.7

Additional context:
YoloSharp.Gpu 6.0.7 works fine
CUDA version and drivers are installed correctly
Issue seems related to QuickGelu node or GPU kernel compilation

Request:
Please fix GPU support for GTX 1080 Ti in a future release.

Thank you for maintaining the library!

[dme-compunet]

The only logical difference that could explain this issue is the version of Microsoft.ML.OnnxRuntime.Gpu used by the package.

In YoloSharp.Gpu 6.0.7, the dependency version is 1.22.1, while in 6.1.0 it was updated to 1.23.2.

Since 6.0.7 works correctly on your GPU, the problem is most likely related to a change in Microsoft.ML.OnnxRuntime.Gpu rather than in this library itself.

Unfortunately, I don't have much hands-on experience with CUDA or GPU-specific issues, and the error appears to originate from the underlying ONNX Runtime GPU implementation.

For this reason, it would probably be more appropriate to open an issue in the official ONNX Runtime repository so the maintainers there can investigate the CUDA kernel compatibility with the GTX 1080 Ti.

[Alex-code24]

Hello,

I opened an issue in the ONNX Runtime repository regarding the GPU crash:
microsoft/onnxruntime#27588

They replied that Pascal architecture (GTX 1080 Ti) is supported and asked for a minimal reproduction project.

The crash happens when using YoloSharp.Gpu 6.1.0 which internally uses Microsoft.ML.OnnxRuntime.Gpu 1.23.2. The same code works correctly with YoloSharp.Gpu 6.0.7 which uses OnnxRuntime.Gpu 1.22.1.

Since the only change between the working and crashing version is the ONNX Runtime GPU package, it seems the issue might be related to that update or possibly how YoloSharp initializes the ONNX Runtime GPU provider.

Unfortunately I cannot share my full project, but the crash happens during normal inference on a GTX 1080 Ti.

Thank you.

[dme-compunet]

Does the error occur only during inference, or while loading the predictor?