Building off of #3007 it would be good to further investigate CuPy, especially after they added support for NEP-18 in cupy/cupy#1650 .

To investigate this further, I recommend the following steps

1. Install cupy with that change installed (depending on when the last release was, this might only be in development versions)

2. install dask

3. Allocate a cupy array of random data

   import cupy
   x = cupy.random.random((1000, 1000))

4. Split that array into a dask array

   import dask.array as da
   d = da.from_array(x, chunks=(100, 100), asarray=False)

5. Start trying out Numpy syntax, and see what breaks

   x.sum().compute()
   x.mean().compute()
   (x + x.T).sum(axis=1).compute()
   ...  # there are many other things that we could do here
   u, s, v = da.linalg.svd(x)
   s.compute()

Probably after this we will learn about many flaws either in Dask Array, CuPy, or both.

cc @zronaghi @pentschev