Arch Linux

Arsakes

Member

Registered: 2011-09-17

Posts: 27

Tensorflow with CUDA doesn't work for version 2.4

Whenever I attempt to use tensorflow with GPU support i get (cpu operations work fine)

2021-03-15 01:08:23.907607: F ./tensorflow/core/kernels/random_op_gpu.h:244] Non-OK-status: GpuLaunchKernel(FillPhiloxRandomKernelLaunch<Distribution>, num_blocks, block_size, 0, d.stream(), key, counter, gen, data, size, dist) status: Internal: no kernel image is available for execution on the device

Is it only me?

System
kernel 5.10.23-1-lts  x86_64 GNU/Linux

GPU
Nvidia Quadro M1000M
NVIDIA-SMI 460.56       Driver Version: 460.56       CUDA Version: 11.2 

Packages I'm using:
- community/python-tensorflow-opt-cuda 2.4.1-6
- community/tensorflow-opt-cuda 2.4.1-6
- community/cudnn 8.1.0.77-1
- community/cuda 11.2.1-4
- python (3.9)

EDIT:

I've checked something:
The tensorflow-2.4.1 installed by pip on python38 (from  AUR) works  and that is using the same cuda and cudnn packages.

Last edited by Arsakes (2021-03-15 01:04:47)

pacman41

Member

Registered: 2019-12-01

Posts: 24

Re: Tensorflow with CUDA doesn't work for version 2.4

Not just you I am having the same problem seems to be a missing library.

error while loading shared libraries: libprotobuf.so.23: cannot open shared object file: No such file or directory

I Downgraded

protobuf-3.12.4-3 
protobuf-c-1.3.3-2 
python-protobuf-3.12.4-3

Last edited by pacman41 (2021-03-15 12:10:25)