my gpu is gt940m
cuda vision is 8.0
cudnn visions 5.1
and i run the examples,it works and the execution times is similar as you.
but for the recurrent networks, Coriander is around 20 times slower than using CUDA .not four
and even slower than using cpu.
is this normal ?
my gpu is gt940m cuda vision is 8.0 cudnn visions 5.1 and i run the examples,it works and the execution times is similar as you. but for the recurrent networks, Coriander is around 20 times slower than using CUDA .not four and even slower than using cpu. is this normal ?