create a C benchmark with cublas

MatthieuCourbariaux / BinaryNet

Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1

BSD 3-Clause "New" or "Revised" License

1.04k stars 346 forks source link

create a C benchmark with cublas #1

Closed ppwwyyxx closed 8 years ago

ppwwyyxx commented 8 years ago

Hi, Your work is great! I did some benchmark in C++ and it shows that xnor is 23x faster than your baseline gemm, and 3.3x faster than nvidia cublas gemm, on GTX980.

Also, I think comparing with Theano doesn't make a lot of sense because it looks like everything got slow down in Theano. cublas would be a good baseline to compare with, since it is the fastest available gemm implementation.

MatthieuCourbariaux commented 8 years ago

Hi Yuxin,

I tried your C benchmark and it works great, thank you very much! I think you are perfectly right, it makes much more sense to compare our kernel with cublas. You just earned a place in our article acknowledgements :)