it varies a lot these days, since theres are many different forward clustering methods. mine specifically does not effect draw calls:
renderdoc shows a few dozen drawelements on the left, to draw the world from one view (somewhere amongst them theres a switch between two vertex buffers).
my lights are stored as a global array per world in a texture buffer, and the spacial index is the 3d texture on input 4. the reason for the limit of 1000 lights is that those indices are 10 bits. the textures are calculated on the CPU because i'm not that high tech yet. so the result is that the pixel shader can look up by himself what lights it needs to shade with, which allows render big massive chunks of the world at once and not really have to worry about draw calls or culling far away lights