It's strange.C++ predicts much more slowly than python predicts

[QueensGambit]

Hi @DarkWings,
I know you posted this a while ago.
How are you measuring the runtime in C++, what batch-size do you use and are you using CPU or GPU?
For me the runtime was similar so far on MXNET Intel MKL, CPU, batch-size 1.

Related topic:

Run time is different between python and c++?

I built mxnet by source using MKLDNN and Openblas. And installed python-mxnet by pip install mxnet-cu100. Then I test inference runtime with the same model and same image. My CPU is 4 cores Intel® Core™ i5-7500 CPU @ 3.40GHz. I tested by clock(). python-mxnet : forward runtime 0.0045s c++ : forward runtime 0.7584s My CPP code refered to image-classification. I am surprised by the order of magnitude difference. MXPredGetOutput(pred_hnd, output_index, &(data[0]), static_cast<mx_uint>(siz…