Cuda error launch failed

Description

I use Matlab and its Experiment Manager to train neural networks. Everything was fine but in one session there was an error. Afterwards the gpu is not usable for deep learning training. See details below.

Environment

Matlab 2021a

GPU Type: RTX 3080
Nvidia Driver Version: 471.86
CUDA Version: 11.4
CUDNN Version: 11.4
Operating System + Version: Windows 10 home 20H2
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

Steps To Reproduce

  1. I ran Experiment Manager in Matlab without a problem for days. Then, in one session, an "Out of Memory: error occurred.
  2. Rerun the same experiment, the “Cuda_error_launch_failed” error was shown and cannot continue.
  3. All previous working Matlab deep learning programs stop working with the same “Cuda_error_launch_failed” error.
  4. Delete all files from the OS drive. Reinstalled OS Matlab, RTX 3080 driver, Cuda toolkit, CuDNN. The “Cuda_error_launch_failed” error still occurs. Repeat the installation multiple times, but the problem cannot be solved.

Hi @fz0001,

This forum talks more about updates and issues related to TensorRT. We recommend you to please post your concern on CUDA related forum to get better help.

Thank you.

Thank you for your response.

Good morning,

It seems that you suggest installing TensorRT. After reading the installation guide, it seems that it is not related to Matlab. And I did not need that for Matlab before the error occurred.

Is there some other installation guide that you want to suggest?

Thanks,
Frank.

Hi,

Hope following link may help you.
https://www.mathworks.com/help/gpucoder/ug/tensorrt-target.html

Thank you.

Thanks! Will try.