how to find the difference between the original version and this one about caffe HOT 3 CLOSED

If by "original copy" you mean BVLC/caffe and by "this one" you mean intel/caffe then you could
check performance by using "time" opetion of caffe tool to see average performance of diffrent
primitives executed for given model eg. for bvlc_reference_caffenet (this require setting up ImageNet)

./build/tools/caffe time --model models/bvlc_reference_caffenet/train_val.prototxt --iterations=100 --engine CAFFE

./build/tools/caffe time --model models/bvlc_reference_caffenet/train_val.prototxt --iterations=100 --engine MKL2017

And do similar commandline for bvlc/caffe:
./build/tools/caffe time --model models/bvlc_reference_caffenet/train_val.prototxt --iterations=100

You can start with smaller model like mnist or cifar that is easier to set up (dataset is to be downloaded via provided scripts)

Hope this will help

from caffe.

Sorry for the late response, actually, I do believe that Intel caffe will have much better performance over the original one on Intel CPUs, however, what I want, is to learn how to improve software performance on OpenCL/Intel, maybe not on caffe, but Intel/caffe will be a good example for me to learn. So, I want to compare the original one and the Intel/caffe, to find out what is the difference, then I could learn a lot on it.

I have plan to port other deep learning softwares from Cuda to Intel/openCL, as I have interest on Intel/opencl/FPGAs, Intel/caffe could be a good reference.

Thank you.

from caffe.

The best resource for OpenCL learning with Caffe will be the opencl branch on BVLC Caffe:
https://github.com/BVLC/caffe/tree/opencl

from caffe.