Tensorflow with CUDA

Support Graphics & Display
1

I use Manjaro stable branch and keep my installation up to date whenever updates are available. The issue I’m having with Tensorflow is that it cannot use the GPU in my computer, but Pytorch works fine. See below:

$ python
Python 3.14.5 (main, May 10 2026, 18:26:20) [GCC 16.1.1 20260430] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>> torch.cuda.get_device_name(0)
'NVIDIA GeForce RTX 5070 Ti'
>>>

>>> import tensorflow as tf
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1781885770.635198   12852 port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
I0000 00:00:1781885770.659255   12852 cpu_feature_guard.cc:227] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX_VNNI AVX_VNNI_INT8 AVX_NE_CONVERT, in other operations, rebuild TensorFlow with the appropriate compiler flags.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1781885771.197680   12852 port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
I0000 00:00:1781885771.197825   12852 cudart_stub.cc:31] Could not find cuda drivers on your machine, GPU will not be used.
>>> tf.config.list_physical_devices('GPU')
W0000 00:00:1781885796.360395   12852 gpu_device.cc:2365] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
[]
>>>

According to pacman, the latest cuda package was built on Wed 06 May 2026, and python-pytorch-opt-cuda Fri 15 May 2026. However, python-tensorflow-opt-cuda was built on Wed 15 Apr 2026, before the latest cuda was released. I believe the issue with Tensorflow is because it was built against the previous version of CUDA, thus the Could not find cuda drivers... and Cannot dlopen some GPU libraries messages.

I’d like to ask the Manjaro team to rebuild and release both Tensorflow and Pytorch packages (and maybe other packages that use CUDA) at the same time a new version of CUDA is to be released.

Thanks for reading this.

2

That sounds like a good idea. In the mean time, you can try switching to unstable branch, where cuda is a newer version.

3

tensorflow has not been built on on the system cuda for awhile now. See:

Either way, cuda 13.2.1-2 & 13.2.1-3 were rebuilds to change build dependencies, not a new upstream release. 13.2.1-1 was built on April 13th. There’s nothing to rebuild as far as I can see.

As @Teo mentioned above, you can try switching branches to see if newer packages make any difference.

4

Thank you both for the reply.

Didn’t know that since I only use Pytorch till now, thanks for the info (I almost tried to build Tensorflow myself :downcast_face_with_sweat:). I’ll try the packages in the unstable branch.