I’m on step 3 of section 3.2 of this installation guide. Installation Guide :: NVIDIA Deep Learning cuDNN Documentation . My goal is to be able to load the dynamic library cudnn64_8.dll . The instructions say to copy some files into some destination folders. I have the files. But I don’t have the destination folders. Should I create the destination folders such as C:\Program Files\NVIDIA\CUDNN\v8.x\bin or was I supposed to get these folders from somewhere else? To be clear, I don’t even have C:\Program Files\NVIDIA\. I only have C:\Program Files\NVIDIA Corporation and C:\Program Files\NVIDIA GPU Computing Toolkit.
I’m running some TensorFlow code and I get this warning.
2022-10-02 22:39:44.797031: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found
What should I do?
Hi @urkchar ,
Which version of cudnn you are using?
I believe this post should be able to help you.
Thanks!
I solved the issue using miniconda3 to install TensorFlow and most of its requirements. This topic may be closed now. Thank you.
I have this exact same problem. What should I do?
I have the same problem. I have installed CUDA Toolkit 11.8. Am trying to install cuDNN 8.4.1.50. Can downlod the file. The instructions ask you to unzip it in the . What Do I create a new folder called NVIDIA|cuDNN\v.8.4.1 in my program files? The instructions are extremely convoluted. I wish someone would post clear instructions. I will try using miconda3. However, do I not need this to be installed and have an environment path for me to use it in Tensorflow 2.10? Thanks
you have to put environment variables in windows.
First create for example a tools folder in C.
Put in cudnn and tensorrt folder separately.
In environment variables windows put in path:
cudnn/lib
cudnn/include
and tensorrt the same
Came here with the same question. Sadly I don’t see an answer that is relevant. If I figure it out, I’ll post the answer. The instructions need to be updated.
I figured out the answer, there is a new name for the folder.
For me, instead of
Copy bin\cudnn*.dll to C:\Program Files\NVIDIA\CUDNN\v8.x\bin.
I had to do
Copy bin\cudnn*.dll to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDNN\v12.x\bin.
This also applies to the ‘include’ and ‘lib’ folder.
If people are curious about my version, I’m on 12.1
I had the same problem and no answer to be found.
So to install cudnn I just use what I have, I have this in my machine “C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin” so I used it and it seems like working.
I said it was working as I was able to compile get the darknet.exe based on the steps below
Before I install cudnn, I used to get the error “Could NOT find CUDNN (missing: CUDNN_INCLUDE_DIR CUDNN_LIBRARY)” in the cMake logs.
I have the same problem,I don’t have the floder “C:/NVIDIA”,so sad.It waste a lot of time
Same problem here, don’t have C:\Program Files\NVIDIA\CUDNN\ or C:\Program Files\NVIDIA GPU Computing Toolkit\CUDNN\v12.x. OMG Nvidia just please double-check the installation guide before publishing.
Configuring these drivers is utter torture. All support files lead to oudated instructions, with specify versions THAT DO NOT MATCH what folder structure they actually use. I found an older archived PDF, that for whatever reason, matches the file paths you’ll see, with the version that are usable in popular neural network packages. I don’t yet know if it will work, but the folder names actually match.
- Copy the following files into the CUDA Toolkit directory.
a). Copy \cuda\bin\cudnn*.dll
to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\bin.
b). Copy \cuda\include\cudnn*.h
to C:\Program Files\NVIDIA GPUComputing Toolkit\CUDA\vx.x\include.
c). Copy \cuda\lib\x64\cudnn*.lib
to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\lib\x64.
HELLO NVIDIA. Can you update your support documentation? We’re suffering here as customers.