onnxruntime_providers_cuda.dll cannot be loaded due to "Can't find dependent libraries" under Windows 10 environment using Java

[felixchan68]

Describe the issue

Try to execute the Java code under Windows 10 environment with GPU supported,

onnxruntime_providers_cuda.dll: Can't find dependent libraries will be shown during the code execution. Detail reported under the "To reproduce" section.

The native library "onnxruntime_providers_cuda.dll" extracted from the path "ai/onnxruntime/native/win-x64/" within the onnxruntime_gpu-1.15.1.jar are included within the IDE Eclipse

To reproduce

public static void main (String [] args) {
System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
System.loadLibrary("onnxruntime_providers_cuda"); // <-- will show "Can't find dependent libraries" during runtime

OrtEnvironment environment = OrtEnvironment.getEnvironment();
OrtSession.SessionOptions sessionOptions = new OrtSession.SessionOptions();

sessionOptions.setOptimizationLevel(OrtSession.SessionOptions.OptLevel.ALL_OPT);
sessionOptions.addCPU(false);
sessionOptions.addCUDA(0); // <-- will show "LoadLibrary failed with error 126 " when trying to load onnxruntime_providers_cuda.dll" during runtime

}

Urgency

Yes project blocked if no GPU supported

Platform

Windows

OS Version

10

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.15.1

ONNX Runtime API

Java

Architecture

X64

Execution Provider

CUDA

Execution Provider Library Version

Build cuda_12.2.r12.2/compiler.32965470_0

[skottmckay]

Have you installed cuDNN and its dependencies as those are required? https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html

CUDA 12 isn't officially supported yet. #16675 (comment)

However, others have successfully used ORT with CUDA 12. #16713 (comment)

[mszhanyi]

Could you please check C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v{cu_version}\extras\CUPTI\lib64 in PATH
{cu_version} stands for the CUDA you installed, e.g. 11.8