caffe训练时总是会占用0号GPU卡

 Sometimes, we will change the default GPU 0  to others when  we are training deep learning models using Caffe, however, we also find that gpu 0 is always used (~100 MBs) 

.To solver this, we can simply change the order of the GPU setting language: 

caffe.set_mode_gpu()

caffe.set_device(2)

=====>

caffe.set_device(2)

caffe.set_mode_gpu()