[Build] CUDA_PATH is set but CUDA wasn't able to be loaded (Windows 11, onnxruntime-gpu 1.13.1, CUDA 11.8)

[Ltbltbltbltb]

Describe the issue

I am trying to use the ONNX Runtime CUDA Execution Provider with a GeForce RTX 4060 Ti on Windows 11, but I am consistently getting the error "CUDA_PATH is set but CUDA wasn't able to be loaded". I have CUDA Toolkit 11.8 and cuDNN 8.9.2 installed, and I am using onnxruntime-gpu version 1.13.1.

I have tried the following troubleshooting steps:

• Reinstalling CUDA Toolkit 11.8 and cuDNN 8.9.2
• Reinstalling onnxruntime-gpu 1.13.1 with the --no-cache-dir flag
• Verifying that the CUDA_PATH, Path, INCLUDE, and LIB environment variables are set correctly
• Running the deviceQuery.exe and bandwidthTest.exe CUDA samples, which both passed
• Running the script and the Deep-Live-Cam application as administrator
• Disabling my antivirus temporarily
• Creating an onnxruntime_config.json file to force the use of the CUDA Execution Provider
• Checking NVIDIA Control Panel settings to ensure that my GPU is selected for CUDA

Despite these efforts, the error persists. I am unable to generate the ONNX Runtime debug log. I am using Python 3.10.9 and Visual Studio Build Tools 2022.

I have attached the output of the onnx_test.py script and the Deep-Live-Cam application, as well as screenshots of my environment variables.

I would greatly appreciate any assistance in resolving this issue.

Urgency

No response

Target platform

Windows

Build script

import onnxruntime as ort
import json

with open('onnxruntime_config.json', 'r') as f:
config = json.load(f)

try:
ort_session = ort.InferenceSession(r"C:\Users\lucas\miniconda3\envs\trocarosto\Deep-Live-Cam\models\inswapper_128_fp16.onnx", providers=config['providers'])
print("Sessão ONNX Runtime com CUDA criada com sucesso!")
except Exception as e:
print(f"Erro ao criar sessão ONNX Runtime com CUDA: {e}")

Error / output

(base) C:\Windows\System32>conda activate trocarosto

(trocarosto) C:\Windows\System32>cd C:\Users\lucas\miniconda3\envs\trocarosto\Deep-Live-Cam

(trocarosto) C:\Users\lucas\miniconda3\envs\trocarosto\Deep-Live-Cam>python onnx_test.py
EP Error using [{'name': 'CUDAExecutionProvider', 'device_id': 0}]
Falling back to ['CUDAExecutionProvider', 'CPUExecutionProvider'] and retrying.
2 0 2 4 - 0 8 - 1 4 1 5 : 2 9 : 4 5 . 0 7 5 7 7 1 7 [ W : o n n x r u n t i m e : , g r a p h . c c : 3 4 8 7 o n n x r u n t i m e : : G r a p h : : C l e a n U n u s e d I n i t i a l i z e r s A n d N o d e A r g s ] R e m o v i n g i n i t i a l i z e r ' b u f f 2 f s ' . I t i s n o t u s e d b y a n y n o d e a n d s h o u l d b e r e m o v e d f r o m t h e m o d e l .
2 0 2 4 - 0 8 - 1 4 1 5 : 2 9 : 4 5 . 1 0 9 4 0 4 5 [ E : o n n x r u n t i m e : D e f a u l t , p r o v i d e r _ b r i d g e _ o r t . c c : 1 2 6 6 o n n x r u n t i m e : : T r y G e t P r o v i d e r I n f o _ C U D A ] D : \ a \ _ w o r k \ 1 \ s \ o n n x r u n t i m e \ c o r e \ s e s s i o n \ p r o v i d e r _ b r i d g e _ o r t . c c : 1 0 6 9 o n n x r u n t i m e : : P r o v i d e r L i b r a r y : : G e t [ O N N X R u n t i m e E r r o r ] : 1 : F A I L : L o a d L i b r a r y f a i l e d w i t h e r r o r 1 2 6 " " w h e n t r y i n g t o l o a d " C : \ U s e r s \ l u c a s \ m i n i c o n d a 3 \ e n v s \ t r o c a r o s t o \ L i b \ s i t e - p a c k a g e s \ o n n x r u n t i m e \ c a p i \ o n n x r u n t i m e _ p r o v i d e r s _ c u d a . d l l "

Erro ao criar sessão ONNX Runtime com CUDA: D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:574 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. Please install the correct version of CUDA and cuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/reference/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

(trocarosto) C:\Users\lucas\miniconda3\envs\trocarosto\Deep-Live-Cam>

Visual Studio Version

Visual Studio Build Tools 2022

GCC / Compiler Version

No response

[tianleiwu]

You will need set PATH environment variable to directories containing cuda, cudnn and VC Runtime DLLs.
Example is like
set PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\libnvvp;C:\nvidia\cudnn-windows-x86_64-8.9.6.50_cuda11-archive\bin;C:\Windows\system32

[Ltbltbltbltb]

Thank you very much my Man! You're a Genius!