search

BVLC / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
34.04k stars 18.7k forks source link

cuda8.0 + cudnn7.1 consumes GPU memory and memory more than cuda8.0 + cudnn5.1 #6412

Open tzhang2014 opened 6 years ago

tzhang2014 commented 6 years ago

win7 SP1 ,GTX1080 , VS2013, cuda8.0 + cudnn7.1 consumes GPU memory and memory more than cuda8.0 + cudnn5.1 to test caffeNet , Why? THX

Noiredd commented 6 years ago

This could be due to how cuDNN treats grouped convolutions, CaffeNet has a few of those. But what kind of difference are we talking about here?

ShawKai666 commented 6 years ago

@Noiredd CuDNN 7 grouped convolution is not fast enough on MobileNet，why？