Skip to first unread message
Abhinav Kumar Gupta
Oct 21, 2015, 4:05:51 AM10/21/15
to Caffe Users
Hi all,
What are the various ways to improve efficiency of Caffe Model in making predictions using only CPUs ?
I have read on forums that using Intel MKL can help in speeding up predictions ? Is that true ?
Thanks,
Abhinav
What are the various ways to improve efficiency of Caffe Model in making predictions using only CPUs ?
I have read on forums that using Intel MKL can help in speeding up predictions ? Is that true ?
Thanks,
Abhinav
Ronghang Hu
Oct 21, 2015, 1:08:03 PM10/21/15
to Caffe Users
Intel MKL should be much faster than the default Atlas, and you can use a 30-day trial version if you only need it temporarily. Or you can get a free education license if you are in a university.
OpenBLAS should also be faster than Altas, although may not be as fast as MKL.
zhang wang
Oct 22, 2015, 7:28:58 AM10/22/15
to Caffe Users
2 papers may be helpful for you.
V. Vanhoucke, A. Senior, and M. Z. Mao, “Improving the
speed of neural networks on CPUs,” in Deep Learning and
Unsupervised Feature Learning Workshop, NIPS 2011, 2011
Accelerating Very Deep Convolutional Networks for Classification and Detection
在 2015年10月21日星期三 UTC+8下午4:05:51，Abhinav Kumar Gupta写道：
Accelerating Very Deep Convolutional Networks for Classification and Detection
在 2015年10月21日星期三 UTC+8下午4:05:51，Abhinav Kumar Gupta写道：
George Henrique Silva
Nov 2, 2016, 8:18:12 PM11/2/16
to Caffe Users
I would recommend to use the Intel fork of Caffe:
It has more than the MKL optimizations and Intel keeps it in synch with the original Caffe master branch.
__
George H. Silva
Community Manager for Machine Learning
Intel Corporation
On Wednesday, October 21, 2015 at 1:05:51 AM UTC-7, Abhinav Kumar Gupta wrote:
Reply all
Reply to author
Forward
0 new messages