0

I'm trying to use Ubuntu 22.04 for my CNN project. For that I'm using Jupyter Notebook. When I'm running my code in the notebook suddenly the kernel crashes.

After following all the instructions available in nvidia such as https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-linux, etc. and other platforms, I'm still not able to work properly with this combination. Is it possible to update the driver or reduce the CUDA version? I have tried multiple solutions, but nothing is working out.

Jupyter Notebook error:

Jupyter notebook error

NVIDIA driver information:

Nvidia driver information

karel
  • 114,770
Spruha Pingale
  • 1

1 Answers1

2

The compute capability of your GT 610 is 2.1, and that limits the CUDA release you can run to 8.x. You may check this at the Nvidia site, but successfully ran CUDA 8.0 with a Quadro 1000m whose CC was also 2.1, and set up cudnn and tensorflow for that.

Old versions of CUDA are available from Nvidia. Check this site for installing CUDA into your own directory instead of any system area,(CUDA 10.2 - different installation paths) .

Each CUDA installation will have its own ...cuda/bin directory for executables (like gcc) and ...cuda/lib for libraries. These CUDA executables and libraries may be quite old, much older than their current system versions in /usr/bin and /usr/lib. When doing CUDA work, a script putting the ...cuda directories at the beginning of PATH and LD_LIBRARY_PATH will allow the CUDA builds to pick them up. Other system programs (e.g. update) avoid using the old CUDA versions since they will not be present in the system areas specified in the default PATH or LD_LIBRARY_PATH.

ubfan1
  • 17,838