Error when installing tensorflow 2.19 with GPU support

Hi. I tried to install the newest tensorflow version (2.19) with GPU support. However after multiple tries it still won’t work and throws the following error messages:

2025-06-02 20:23:33.262356: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1748888613.270707  104143 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1748888613.273217  104143 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

I followed the exact installation procedure that was shown in the guide and (I believe) installed CUDA Toolkit and cuDNN correctly. Any help would be much appreciated. :slight_smile:

System specs:

  • Linux Version: Ubuntu 22.04.5 LTS
  • NVIDIA RTX 3500 Ada Generation
  • NVIDIA Driver Version: 555.42.06
  • CUDA Toolkit Version: 12.5.82
  • cuDNN Version: 9.3.0.75
  • Python Version: 3.12.10
1 Like

After several weeks of upgrading and downgrading libraries, compiling tensorflow on my local machine and other things, I finally tried other combinations of cuDNN, CUDA and Tensorflow. Finally, I found a combination that works for me:

  • CUDA 12.5
  • cuDNN 8.9
  • Tensor Flow 2.16.1

I don’t know why this specific combination works and I don’t give
But as I actually wanted to upgrade to the newest version of tensorflow, this solution is suboptimal.

Also a lack of answers here in this forum (or for that matter anywhere else in the internet) I have the slight suspicion, that the “pretested” configuratons found here do not really work and are not really pretested.
A more user friendly way to install tensorflow with gpu support is therefore highly desirable.

2 Likes