The procedure entry point could not be located in the DLL Library

[Subash-Chandra]

The prodecure entry point ?destroyTensorDescriptorEx@ops@cudnn@YA?AW4cudnnStatus_t@@PEAPEAPEAUcudnnTensorStruct@@@z could not be located in the dynamic link library C\sdwebui\venv\lib\site-packages\nvidia\cudnn\bin\cudnn_adv_infer64_8.dll.

I get this for 4 diferent DLL's when I try to run automatic1111 after installing this extension. I've installed CUDA and CUDNN.

[concrete13377]

I have the same on win11

[plugcrypt]

Same issue using updated version of A1111, Win10, RTX4090 / 545.92 Nvidia drivers
Replacing dlls from https://developer.nvidia.com/rdp/cudnn-archive didn't work. Still got these errors

[Gyramuur]

I get about four of these popups myself before it finally launches. It doesn't seem to affect the functionality in any way but defo annoying. Also should be noted that these popups occur not only on the version of WebUI that I installed the extension on, but on EVERY SINGLE VERSION of Stable Diffusion across my computer, even backup versions from months and months ago that I've left untouched.

So whatever it is from this extension that's caused this, it's system-wide, lol.

[contentis]

Please follow these instructions: #27 (comment)

[MasterOfPhantom]

Please follow these instructions: #27 (comment)

still not working for me

[riperbot]

I found a simple solution. It worked for me and stopped showing error windows at startup. Just delete these files in the folder "c:\sdwebui\venv\lib\site-packages\nvidia\cudnn\bin":

• cudnn_adv_infer64_8.dll
• cudnn_adv_train64_8.dll
• cudnn_cnn_infer64_8.dll
• cudnn_cnn_train64_8.dll

Because they are also located in the "c:\sdwebui\venv\Lib\site-packages\torch\lib" folder. But I noticed they're not the same size. Everything works as before, just without those annoying error windows.

[darebgd]

@riperbot, your solution helped, but it is rather brute-forcing of stuff to work properly. To do it more gracefully, activate your virtual python environment (venv) which Auto1111 uses (open the command prompt, execute activate.bat script), and type:

pip uninstall nvidia-cudnn-cu11

and

pip uninstall nvidia-cudnn-cu12

TensorRT extension installation kind of does this for you, but still, make sure you check in your venv with:

pip show nvidia-cudnn-cu11 and pip show nvidia-cudnn-cu12, respectively

You might have installed nvidia-cudnn-cu12 without even knowing, so make sure you have no NVIDIA CUDNN packages in your venv where Automatic1111 runs.

Tested on:

• 4080 RTX Super GPU
• Automatic 1111 v1.9.0
• xformers 0.0.25.post1
• pytorch 2.2.2+cu121

[Gihasy]

Thanks @darebgd it solved for me!