Could you please give a detailed training environment configuration?

[Imalne]

When I tried to train the network with cuda 10.1, cudnn 8.0.2 and torch 1.4.1, it always reported a "core dump" error. And when I turned off cudnn the error disappeared but I can't train with batch size as large as 1024. I guess it is a version adaptation problem

[Cliveliew]

When I tried to train the network with cuda 10.1, cudnn 8.0.2 and torch 1.4.1, it always reported a "core dump" error. And when I turned off cudnn the error disappeared but I can't train with batch size as large as 1024. I guess it is a version adaptation problem

i train the network with cuda9.0, torch 1.1.0 and change the batch_size to 64, finally it works