How can I speed up the training process using GPU or multi thread on CPU?

Question

How can I speed up the training process using GPU or multi thread on CPU?

striversist opened this issue 8 years ago · comments

Even if I build the project success using command "scons -j4 gpu=1", it seems the clstmocrtrain still use CPU to train which is realllllly slow.
So is there any option to turn GPU training on? Was the project dead?

Tom · Answer 1 · Fri Oct 28 2016 05:34:08 GMT+0800 (China Standard Time)

clstm+Eigen can perform the computations on the GPU in principle, but that code hasn't really been tested much, and it turns out there are now better ways to handle LSTM on GPU. I'm working on a new project for high-performance GPU-based OCR.