Cuda Error : RuntimeError: CUDNN_STATUS_EXECUTION_FAILED

Not sure if my experience would help someone else. My error was caused due to incompatible torch version in my docker environment.

Execute torch.version.cuda from python shell to check if that is compatible with your CUDA versions in nvcc -V or nvidia-smi in the bash shell.

For my case, the torch version was too old that it only supported CUDA 10.1 but my system was using 11.3.

Here is a more detailed post of the same error being solved by upgrading the torch version. Cheers.

1 Like