W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found

[marshallia]

Please make sure that this is a build/installation issue. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:build_template

System information

• OS Platform and Distribution Windows 10 x64
• TensorFlow installed using pip
• TensorFlow version 2.5.0-dev20201028
• Python version: Python 3.8.6 (tags/v3.8.6:db45529, Sep 23 2020, 15:52:53) [MSC v.1927 64 bit (AMD64)] on win32
• Installed using virtualenv? pip? conda?: pip
• CUDA/cuDNN version: cuda 11.1/cudnn 80.04
• GPU model and memory: NVIDIA RTX3080

Hi,
I got this issues when I am trying to import tensorflow from command line.
Currently I am using NVIDIA RTX3080, with cuda 11.1, cudnn 8.04, python 64 bit, and nvidia driver 456.43.
I follow the instruction on https://www.tensorflow.org/install, and install tf-nightly 2.5.0.dev and tf-nightly-gpu 2.5.0.dev.
I also have try the solution on this discussion but it didn't work me #43193.
But I am still getting this error

[king398]

I think You should check if you have added all the cuda enviroment path variable. Type this in you command prompt
PATH

[Saduf2019]

@marshallia
please refer to these issues with same error and let us know: link, link1, #43193

Have you set in PATH
SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11\bin;%PATH%

[marshallia]

@marshallia
please refer to these issues with same error and let us know: link, link1, #43193

Have you set in PATH
SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11\bin;%PATH%

Hi, Saduf.
I have give up on installing tensorflow on windows. Instead of using Windows I use Ubuntu and install cuda 11.1 package that contains nvidia driver inside it. Then install tensorflow nightly. And it is work for me

Cuda 11.1
Nvidia driver 455.32
Tensorflow 2.5.0-dev
python 3.8.5
Ubuntu 20.04

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No

[CostanzoPablo]

I have the same problem and the solution is:

1. Uninstall all drivers that says "nvidia" from Uninstall programs (ALL, Cuda and gpu driver)
2. Install last nvidia drivers for your GPU https://uk.download.nvidia.com/GFE/GFEClient/3.21.0.36/GeForce_Experience_v3.21.0.36.exe
3. Install cuda 11.2 https://developer.download.nvidia.com/compute/cuda/11.2.1/local_installers/cuda_11.2.1_461.09_win10.exe
4. Install Anaconda https://repo.anaconda.com/archive/Anaconda3-2020.11-Windows-x86_64.exe
5. conda create -n cudaenv python=3.6
6. conda activate cudaenv
7. pip install tensorflow python-dotenv
8. Create a.py with this: import tensorflow as tf
9. run it: python a.py
   Output:
   xxx: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library cudart64_110.dll

[pythonic2020]

@CostanzoPablo, thank you very much for the above post. I finally got tf 2.4.1 to install in a conda-forge environment with GPU support (CUDA 11.3, Python 3.8.8)! Where did you find out that "python-dotenv" should be installed? I haven't seen that package mentioned anywhere on tf website, or anywhere else, with regard to tf install.

However, the above post leaves out many important steps that are required to get tf working with GPU support.

[sprash]

@CostanzoPablo That worked for me! Thanks for posting this solution.