0

When I'm trying to run my Tensorflow code inside my conda environment myEnv001, I got the error message dlerror: cudart64_110.dll not found. I'm aware that my error is similar to a previous post, which says I didn't install CUDA.

I have CUDA 10.1 installed, but anyway I tried to re-install CUDA 10.1 for myEnv001. I ran below at my terminal, and CUDA 10.1 is installed successfully in myEnv001

> conda activate myEnv001
(myEnv001) > conda install -c anaconda cudatoolkit=10.1

But when I tried to run my Tensorflow code I still have the same error message... How can I fix it?

My library

  • conda: 4.10.1
  • Tensorflow installed in myEnv001: 2.6.0
  • cudatoolkit installed in myEnv001: 10.1.243
talonmies
  • 70,661
  • 34
  • 192
  • 269
H42
  • 725
  • 2
  • 9
  • 28
  • @talonmies Thanks for your reply. I just installed CUDA 11.0 in `myEnv001`. But then the error message says `dlerror: cusolver64_11.dll not found` (i.e. another file is missing). Any ideas? – H42 Sep 20 '21 at 14:47
  • 1
    No, you need CUDA 11.2, remove that 11.0 and install 11.2. You also need to have cuDNN 8.1. Check [this](https://www.tensorflow.org/install/source_windows#gpu) link, there is a table at the very end of that page. Check that. – Adarsh Wase Sep 20 '21 at 18:39

1 Answers1

1

According to Tested build configurations for TF 2.6, the compatible CUDA is 11.2 and cuDNN is 8.1.

For the benefit of community providing compatibility versions are as shown below

enter image description here

For more details, you can refer windows and Linux/macOS tested build configurations for CPU and GPU.