About cudnn.benchmark for 3D convolution

(Kaiyu Yue) #1

Hi, developers:

Does the cudnn.benchmark not works when using conv-3D to build a network? The input size of model is fixed during training and inference.

The cudnn error is:
RuntimeError: CUDNN_STATUS_INTERNAL_ERROR THCudaCheck FAIL file=/pytorch/torch/lib/THC/generic/THCStorage.c line=184 error=77 : an illegal memory access was encountered terminate called after throwing an instance of 'std::runtime_error' what(): cuda runtime error (77) : an illegal memory access was encountered at /pytorch/torch/lib/THC/generic/THCStorage.c:184 Aborted (core dumped)

Many thanks.


hmmm that’s weird, looks like maybe a CuDNN bug. are you on pytorch v0.3.0?

(Kaiyu Yue) #3

yes! This is the env:

Python 2.7.14 |Anaconda, Inc.| (default, Oct 16 2017, 17:29:19)
> import torch
> torch.__version__