I was thinking that numpy is used to operate data on cpu, but I found that when I created a cupy.array, I can still use numpy module to operate it, and it's still on cuda.
Like this
x = cp.array([[1, 2, 3], [4, 5, 6]])
y = cp.array([[1, 2, 3], [4, 5, 6]])
z = np.add(x,y)
It still works!
I thought that maybe numpy has moved data to cpu
So I wrote this
print(type(z))
print(z.device)
It shows that
<class 'cupy.ndarray'>
<CUDA Device 0>
It's still on cuda. How did numpy make this? I mean, numpy should work on cpu right? So is it cupy working or numpy working?