Originally Posted by Hwgeek
What do you think about Full Scene Ray tracing in Cloud?
When Nvidia could make a GPU that can offer 4K Full Scene Ray tracing?
RT on its own cost a hell of a lot of compute. Nvidia had to resort to hybrid RT for now even with 544 dedicated tensor cores for compute, which are the current bottleneck for RT.
To get decent fps at 4K, they will need twice if not more dedicated cores to do it. Going to 7nm, it will still make the core pretty big. So I don't know if they will be able to pull full RT in 4K without hybrid.
This is why I think AMD won't be able to offer a similar solution, and the reason they hadn't even talked about it now, is because they are still working on their own hybrid solution similar to nvidia, until the compute power arrives.
Full scene RT in cloud will introduce a lot of lag if done in real time remotely. Considering we want 4K/144hz monitors and leverage performance for it, I think it is a nice dream but not very realistic for fast moving games.
Would you like your offline game to force you to be 100% online at a minimum lag as possible in order to really view the game as you like? I think hardware solution is a better solution. Maybe not for consoles, but for PC games, definitely.
Originally Posted by umeng2002
A joke. "Cloud" compute for interactive media is pointless... you might as well just stream the entire game.
2nd gen RDNA clearly won't have all the ray tracing acceleration the RTX line had last year. Quake II RTX is already "full scene" ray traced.
Well if you look at AMD, what might benefit them, is their ability to use their cores for both high level compute and graphics. So if they make their next gen have twice the cores, they can turn half the cores into compute cores through their hardware async compute system.
That means they can run at a 2070-ish 4K performance for general purposes and texture rendering, with a high capacity of compute that could match or surface tensor cores for lighting.
This could potentially give them an edge. But, I don't think they will go to such big cores, so they might move to a similar hybrid solution for now, or accept the huge FPS drop from fully supporting the RT API.