Failing to install cuDNN on several PCs

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Hi all,

I’m having problem installing GPU-acceleration with TensorFlow.

The system is an Asus laptop with an Intel i7-7700HQ (4-core) CPU, 16GB DDR4, NVIDIA GeForce GTX1050 GPU, Windows 10.

I have installed CUDA 10, CUDA 10.1, CUDA 10.2, Python 3.8.5, Pip 20.2.3, Tensorflow 2.3.0. The NVIDIA Driver is 451.67.

Pycharm 2020.2.1 Community Edition works (without GPU support).

The installation paths for CUDA 10, 10.1 and 10.2, and the /bin path in CUDA 10, appear correct, as environmental variables CUDA_PATH, CUDA_PATH_V10_1, CUDA_PATH_V10_2, PATH (for the CUDA10/bin and CUDA10_1/bin directories).

The error is that it can’t find cuDNN64_7.dll.

In the CUDA directory (copied from the cuDNN directory) the file is named cuDNN64_8.dll, so I guess it is a later version.

I assume I have to copy the DLL, H and LIB files into the CUDA_V10_2 installation directory.

How do I solve the problem?

Thanks
Pietro

The code run on Pycharm is:

import tensorflow as tf
print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices(‘GPU’)))

The error messages are:

C:\Users\Pietro\PycharmProjects\ch05\venv\Scripts\python.exe C:/Users/Pietro/PycharmProjects/ch05/main.py

2020-10-07 17:49:18.391010: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cudart64_101.dll

2020-10-07 17:49:30.282990: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library nvcuda.dll

2020-10-07 17:49:31.452213: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties:

pciBusID: 0000:01:00.0 name: GeForce GTX 1050 computeCapability: 6.1

coreClock: 1.493GHz coreCount: 5 deviceMemorySize: 4.00GiB deviceMemoryBandwidth: 104.43GiB/s

2020-10-07 17:49:31.452661: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cudart64_101.dll

2020-10-07 17:49:31.730967: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cublas64_10.dll

2020-10-07 17:49:31.931516: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cufft64_10.dll

2020-10-07 17:49:32.019381: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library curand64_10.dll

2020-10-07 17:49:32.390654: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cusolver64_10.dll

Num GPUs Available: 0

2020-10-07 17:49:32.581313: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library cusparse64_10.dll

2020-10-07 17:49:32.583340: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library ‘cudnn64_7.dll’; dlerror: cudnn64_7.dll not found

2020-10-07 17:49:32.583657: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1753] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at 使用 pip 安装 TensorFlow for how to download and setup the required libraries for your platform.

Skipping registering GPU devices…

Process finished with exit code 0

2

I tried the same on a desktop with a Quadro GPU, same problem.

3

Hi @pietromonsurro,
You may need to install cudnn7.
Tensorflow version which you are using is built against cudnn7.
So it won’t work if you install cudnn8 .

Thanks!

4

Well, I guess it’s the solution, but I downloaded the 10.2 package so I didn’t know if cudnn7 is compatible with my version of CUDA, or I shall downgrade to one that uses cuDNN7.

Thank you very much, I hope it’ll work.
Pietro

5

Hi @pietromonsurro,
Please check the below link

Thanks!