Machine learning for cloud based real time ray tracing?

[Nostremitus 2]

Could a system similar to DLSS function as  a physics and lighting system? Where machine learning develops the lighting and shadowmaps, as well as path traced object movement and collision? Is this the end goal of cloud computing in games?

 

 AMD have said that local real-time raytracing is a stopgap until their cloud solution goes live. 

 

Epic have spoken about incorporating cloud based physics and collision so all players see and interact with the same destructable environments...

 

Both of these seem to be different uses for similar cloud-based solutions that are in development. 

 

Thoughts on how this could work without massive pipeline bottlenecks? 

[igormp]

Yes, but I believe it'd be way to inaccurate to actually be useful. Remember that ML isn't something really precise, it just tries its best.

 

The cloud solution idea I guess is just centralizing all of the calculations in a single place and sending out results to the clients. Given a low enough latency, this should be way better than having every player need to do their own calculations for what's effectively the same scenario shared between all of the players.