Mike just didn’t know how to use all of the controls at the time; everything was undocumented. I did make a modified version of that exact scene, and it looks much better in my opinion.
I was talking with NVIDIA, and they are working to keep up with the main branch while also making the system 100% GPU agnostic. Currently, the Path-Tracer runs on any GPU, including AMD, and even on Linux. The only part that is NVIDIA only is Ray Reconstruction, which is currently used for denoising.
However, they are actively working on a second denoiser that can run on any GPU. They have made it abundantly clear that the goal is to not be NVIDIA only.
Those reflections are insane! The next time someone asks how to make reflections in Godot I’m going to refer them to this thread.
Thats great.
The shadows in that last video are convincing, perhaps the fork will be useful for special builds of a game. People will be able to say “here is the standard version, and here is the super HD version with RTX, take your pick”.
Reflections and shadows definitely benefit the most from Path-Tracing. Indirect illumination is also excellent, but if we’re talking absolute game-changers, SDFGI holds its own for GI. However, in my opinion, reflections fall short, and shadow maps are… well, shadow maps. I don’t need to explain why they suck!
That looks much better. Can it soften directional cast shadows?
Yeah, the shadows are amazing. The best part is that they’re often cheaper than shadow maps, since you can have far more shadow-casting lights at very low cost. Path-Tracing avoids the overhead of shadow mapping, and the resolution is effectively infinite something traditional methods just can’t match. I love PT/RT shadows!
As for special builds, all of this can be toggled in game, so a developer could add a menu where players can choose to use Path-Tracing or rasterization. So yeah, that’s totally possible.
Not at the moment, though it’s being worked on. Right now, rays either hit or don’t hit, but variable penumbras are in development.
How does this thing handle scene geometry? It just chews everything without any culling? What happens if there’s a lot of geometry in the scene?
Basically, anything that’s going to be visible in reflections or indirect illumination needs to be visible. There is still work being done on how exactly occlusion culling and related systems should work, but currently, this renders any objects that are needed for off-screen rendering.