@precedenceguo
why before training faster rcnn, we should export MXNET_CUDNN_AUTOTUNE_DEFAULT=0?
And if I don't export this variable, the terminal will report the cudnn find the algorithem to get best performance in training and testing.
So is cudnn in use when I turn off it?
@precedenceguo why before training faster rcnn, we should export MXNET_CUDNN_AUTOTUNE_DEFAULT=0? And if I don't export this variable, the terminal will report the cudnn find the algorithem to get best performance in training and testing. So is cudnn in use when I turn off it?