1

Related to this post

I have installed TensorFlow (GPU) with anaconda using the installation instructions in here

conda create -n tf-gpu tensorflow-gpu
conda activate tf-gpu

But I realized Anaconda is installing library cudnn version 7.6.5

According to TensorFlow page it should not be installing cudnn 7.4?

This can cause any problem?

double-beep
  • 5,031
  • 17
  • 33
  • 41
Valborsf
  • 11
  • 2
  • "...the supported version for TF 2.0 ( with GPU ) is 7.4" -> Where did you find this? [official TF guide says >= 7.6](https://www.tensorflow.org/install/gpu#software_requirements). The page you link shows only the **tested** build configurations, it doesn't mean you must stick to those. CuDNN 7.6 has been in use for quite a long time now – GPhilo Jan 27 '20 at 13:29
  • Anaconda does their own build of TensorFlow, they can use whatever version of CUDA and cudnn they want. They have been building with cudnn 7.6 for the last 7 months: https://github.com/AnacondaRecipes/tensorflow_recipes/blame/master/tensorflow-base-gpu/conda_build_config.yaml#L16 – William D. Irons Jan 27 '20 at 23:19
  • @GPhilo@William D.Irons Thank for your comments, have been of big help ! – Valborsf Feb 01 '20 at 12:37

0 Answers0