Ndarray Vs Minpy Vs Cupy

I have found another framework called “Minpy” made by the same people(dmlc) who built MXNET. It says Minpy is basically numpy with GPU support. There is another framework called Cupy, with the same idea of numpy with GPU support. How Minpy compares with MXNET’s NDArray and Cupy? And which one is faster? Should I stay with MXNET’s Ndarray or go Cupy or Minpy?

PS: I have not problem with MXNET’s Ndarray but just curious to know.

FYI the community are actually working on a numpy compatible major feature that brings all the goodness of minpy to mxnet: https://github.com/apache/incubator-mxnet/issues/14253
There is not much development for minpy for the past two years. I have not used cupy before so I cannot comment on that.

Thanks for reply. It would be definitely great if we can hybridize control flow written in native python using it else and while loops and for loops.