It really feels like several stages could be coalesced into fewer, and fewer conversion steps - like transforming a reflectance/environmentmap into a dual-paraboloid? Why not just sample reflections directly from the resultant cubemap? Is it really generating all these LUTs *every* frame? That's crazy! While I appreciate that they've pulled off some great rendering for their artists' work, it really seems like a much more efficient and streamlined approach could've been taken. I see a game like DOOM render equally impressive graphics, and wish Cyberpunk's art and scenery could be rendered in there. Granted, idTech doesn't have an open world asset streaming system that could handle the sheer volume of data Cyberpunk entails, but its approach to generating a frame is much cleaner than what's shown here. Still, it's interesting to see how they piled on the effects one-by-one to produce the final rendered frames that the engine does. I've come back to this post a few times over the last year or two and when I saw nobody had commented I thought I'd go ahead and give thanks for your time breaking CP2077's rendering down - and articulating it here too. Thanks!

Wow, that's impressive, the degree of complexity they achieve, certainly not easy to be a professionnal graphics developer