Hi Luciano Resende, could you please elaborate how does this work: “This approach also allows for easy sharing of expensive resources as GPUs”? As far as I know, GPUs are non-sharable resources like RAM, you can’t specify half GPU in kubernetes and in general as well if I am not mistaken.

And another question — with this setup user notebooks are always up and only kernels go up and down on-demand, right? I am checking gateway docs, but still, don’t have it in my mind how notebook + kernel combo works. With pure jupyterhub, we are using webform where the user specifies required parameters, but now sure how it works with single minimal notebook and on-demand kernels