So I've been researching light binning for tiled and clustered rendering. From what I've read, rasterizing light volumes seems to beat async compute in most areas except for:
- Async compute can be overlaid with the depth / shadowmap passes while rasterizing cannot
- Conservative rasterization not fully supported by some HW (GCN most notably)
- Its complex to implement
Dmitry Zhdan has a nice presentation where he laid out how well rasterizing scales as resolution and scene complexity increases. Call of Duty has an impressive z-binning implementation, and MJP used it on The Order and his own DeferredIndexing demo.
But it still seems like there are way more async compute implementations, including Unity with their new HDRP. Is light binning with rasterization really that hard to implement? Or perhaps there are many more games using it that I might have missed?
