How to fix Process finished with exit code -1073741819 (0xC0000005) when using Convolutional layers in pytorch, error on backward()

1

When I use Conv1d or Conv2d layers on pytorch, the process is killed unexpectedly. I am getting the error in the following line:

loss.backward()

My set up:

  • Windows 10
  • cuda 10.2
  • cudnn 7.6.5
  • RTX 2060 Super
  • Nvidia driver 451.67
  • Pycharm 2020.04

Error:

Process finished with exit code -1073741819 (0xC0000005)

In comparison, when I replace the conv layer with a dense layer, the problem doesn't occur.

For more comparison, the same project and the same code was run on Ubuntu 20.04 as well, and it worked quite well.

pycharm
pytorch
conv-neural-network
cudnn
asked on Stack Overflow Jul 23, 2020 by JacquesdeH • edited Jul 24, 2020 by talonmies

1 Answer

0

There seems to be a known bug around this problem that happens with Pytorch on windows, when run on GPU(with CUDA) .

Ensure all params supplied to Conv1d and Conv2d are correct especially padding value. Note that it can have different behaviour with other OS like linux/ubuntu.

And also if you are using Python-3.6 or higher version, it could be this bug. In that case try with Python-3.5


User contributions licensed under CC BY-SA 3.0