How to downgrade to cuda 10.0 in arch linux? - tensorflow

Question

I want to downgrade my cuda 10.1 to cuda 10.0 in arch linux as tensorflow requires cuda 10.0 only.
I installed tensorflow on CUDA 10.1 in arch Linux but I didn't know tensorflow-gpu needs CUDA 10.0 or lesser.
I tried tried symlinking the required 10.0 libraries with new 10.1 libraries but it didn't work.
Whenever I try to import tensorflow in python console error shows up that 'libcublas.so.10.0' not found.
So I found out that I can only run tensorflow on CUDA 10.0 so now I want to downgrade my CUDA. But I have no Idea how to do that in arch Linux.
Every time I run import tensorflow it gives this error.
Traceback (most recent call last):
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/lib/python3.7/imp.py", line 242, in load_module
return load_dynamic(name, filename, file)
File "/usr/lib/python3.7/imp.py", line 342, in load_dynamic
return _load(spec)
ImportError: /usr/lib/libcublas.so.10.0: version `libcublas.so.10.0' not found (required by /home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so)
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/__init__.py", line 24, in <module>
from tensorflow.python import pywrap_tensorflow # pylint: disable=unused-import
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/__init__.py", line 49, in <module>
from tensorflow.python import pywrap_tensorflow
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 74, in <module>
raise ImportError(msg)
ImportError: Traceback (most recent call last):
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/lib/python3.7/imp.py", line 242, in load_module
return load_dynamic(name, filename, file)
File "/usr/lib/python3.7/imp.py", line 342, in load_dynamic
return _load(spec)
ImportError: /usr/lib/libcublas.so.10.0: version `libcublas.so.10.0' not found (required by /home/techievin/.local/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so)
Failed to load the native TensorFlow runtime.
pacman -S cuda=="10.0" is not working.
I searched everywhere but couldn't find any proper answer for arch Linux to downgrade to CUDA 10.0

Answer 1

Download cuda 10.0 package from here: http://archlinux.arkena.net/archive/packages/c/cuda/cuda-10.0.130-2-x86_64.pkg.tar.xz
Install package sudo pacman -U cuda-10.0.130-2-x86_64.pkg.tar.xz

Answer 2

I've solved this issue by using yay. With yay you can install older versions of cuda and cudnn. In my experience this is faster than downloading from archive.org.
Install yay on Arch:
pacman -S --needed git base-devel
git clone https://aur.archlinux.org/yay.git
cd yay
makepkg -si
Search for the version you need:
yay -Ss cuda
yay -Ss cudnn
At the moment of writing this answer I needed to downgrade from cuda 11.2 to 11.0. So I installed specific versions:
pacman -R cuda cudnn
yay -S cuda-11.0 cudnn8-cuda11.0
For me, Tensorflow worked smoothly after downgrading.

Answer 3

I'm facing the same issue and have to re-apply the method I used when I was using my old gtx670:
Downgrade the cuda package
Prevent it from being upgraded
keep in mind that specific Cuda versions need specific nvidia drivers.
Just have a look at the compatibility matrix and eventually prevent also the drivers from upgrading.