mdeff/cnn_graph

InternalError: cudnn PoolBackward launch failed

kuaileyichu opened this issue · 3 comments

kuaileyichu commented

common['p'] cannot be larger than 1, otherwise, InternalError: cudnn PoolBackward launch failed

mdeff commented

I've definitely used it with p > 1. Maybe it's due to a newer version of TensorFlow or cuDNN? Try to downgrade to TF 1.0 and CUDA 8. Or try to run it on the CPU.

kuaileyichu commented

I use TF 1.3. I will try, thanks for your reply.

mdeff commented

You're welcome. Let us know if that solves it. :)