Tensorflow my GPU is not recognized and There are many dll errors

user_18225513 0
Viewed 197

I'm too new with tensorflow and keras, actually I'm trying first to install it correctly. I used Anaconda to make it easier. My question is probabily related to the installation but there might be some erros in script too. Therefore, above, are all steps I made to install tensorflow on anaconda:

#1 - Install with Anaconda prompt as admin

conda create -n tf_gpu python==3.8 conda activate tf_gpu

conda install cudatoolkit=11.2 cudnn=8.1 -c=conda-forge

pip install --upgrade tensorflow-gpu==2.6.0

#2 - test integration

python import tensorflow as tf print(tf.__version__) tf.config.list_physical_devices('GPU')

Test results: I don't know Why none GPU are associated even installing everything, I don't know if the problem is related to this ... (I have a geforce-940mx) ?

Revelant parts of stackTrace:

Many dll erros appeared (dlerror), maybe it's because something wrong with installation:

cufft64_10.dll  not found
cublas64_11.dll  not found
curand64_10.dll   not found
...

#Conclusion

I need help from you guys. I tried many aways to install tensorflow correctly I'm getting headache, everyone said to install it using a new enviroment in anaconda, the steps looks easy, but many issues are happening, why?

#Aditional information

Tested scripts using VsCode

I have Windowns 10

I tried with more than one version using this link: GPU tf installation

anaconda conda keras tensorflow python
1 Answers

Please follow the below steps to have GPU support in your system.

  1. Download the right NVIDIA driver as per your product type (Geforce-940mx)
  2. Download and install cuDNN, CUDA, CUPTI according to installed Tensorflow version in your system. You can check test build configurations here.
  3. Set the path(%PATH%) for each of these in environment variable of your system

For reference,

The detailed instruction steps (with all required links to download NVIDIA driver, CUDA, cuDNN, CUPTI) are mentioned in this Windows setup link to enable GPU support in your system.

Then create a new virtual environment using Anaconda command prompt

conda create -n tf_gpu1 python==3.8 
conda activate tf_gpu1

Install tensorlfow in this (tf_gpu1) virtual environment

pip install tensorflow

Now, check with the below command if tensorflow is able to access the GPU in your system. You can follow the test inegration steps mentioned by you.

python
import tensorflow as tf
tf.__version__      # This code will show the installed tensorflow version
tf.config.list_physical_devices('GPU')   # This will show output in boolean value 'True' or 'False'
user_142906810
Related