Is it possible somehow to use / launch the cudaLaunchCooperativeKernel api with pycuda?
Hoping to achieve sync at grid level with such.
Not at the moment, no.
You have see for yourself here that the driver API version of that functionality doesn't appear anywhere in the current PyCUDA tree. I would opine that adding support would require a fairly major overhaul of the APIs and I am not confident that it will happen soon.