Unknown Soldier
Veteran
Tkumpathenurple
Veteran
That is some very tasteful animation.
Yeah so this looks insane in real life.
Curious...whatever is out of screen is supposed to be culled to save performance. But with ray tracing, reflective surfaces receive information from whatever is out of screen.
Also often the reflections appear to have lower LOD models instead of whats actually on screen
Like in this photo. See trees, shadows on cars, geometry and normal map detail on Spiderman.
So how does this work exactly?
Is there some kind of "cheating" going?
Curious...whatever is out of screen is supposed to be culled to save performance. But with ray tracing, reflective surfaces receive information from whatever is out of screen.
Also often the reflections appear to have lower LOD models instead of whats actually on screen
Like in this photo. See trees, shadows on cars, geometry and normal map detail on Spiderman.
So how does this work exactly?
Is there some kind of "cheating" going?
The BVH structure from which RT information is taken is normally at a lower LOD than the rest of the scene that isn’t reflected. Think of it as a second world in a parallel dimension (at lower LOD) which is only visible in the reflections. You can choose how many objects are in this BVH world, how far you want your reflection to look into etc etc to optimise performance.
This is why the trees have less detail in that example, they are referenced from a lower LOD just for the RT reflex.
That kind of shit could be a hole new kind of adventure/horror game. Make it in vr and will forgive the whole idea has been done in scifi before. Even flash had a season in similar vein.
Unknown Soldier
Veteran
Or you can play The Medium.
But what about the culling? That means the application if culling is also limited as the engine must also render whats out of screen to reflect it on surfaces
That’s why RT is so taxing on the systems
a lot more needs to be rendered as opposed to screen space effects which only need to reference what is on screen at any given time.And speaking of the reflections, I must say that this is by far the most next gen thing I’ve experienced so far.
I just can’t believe how you can see the whole city in the reflection of a building, far far away in the distance, with pedestrians and cars moving around all the way into the distance! I’ve played with that a lot today.
I just can’t believe how you can see the whole city in the reflection of a building, far far away in the distance, with pedestrians and cars moving around all the way into the distance! I’ve played with that a lot today.
D
Deleted member 11852
Guest
Of course you spent all day looking at your reflection!
Of course you spent all day looking at your reflection!
What else is there to look at??
The culling for the main rasterized view can still happen as always. The engine is processing objects and triangles to be sent to the rasterizartion pipeline, and in parallel it has another list of objects and triangles in the RT friendly BVH format for ray tracing. Those can be as similar or different to each other as the dev sees fit. As such, the main view is still culling objects out of the view frustrum as always, but not out of the BVH.
What happened here was an unexpected side effect. It's not about culling stuff out of being rendered, but about culling out of processing their animation. Their animation system skips the animation processing of stuff not seen, which probably leverages the frustum culling list, but they forgot that is not enough once you can see stuff through reflections. So, yeah, RT will make certain performance optimizations in animation processing and skinning obsolete. There is still room to maybe use a rougher aproximation of proper animation for objects that are not in direct view but still possibly visible through RT. This is all stuff for devs to experiment with and develop from now on.
As for RT being more expensive because you can't cull like in rasterization, well, the RT itself is technically a culling system. It steps rays through simplified AABB box aproximations of the scene as much as possible until it has no choice but to test against the final LOD. In a further away future where real time engines are 100% RT, with no rasterization pass, "classic" style culling is not as necessary, the BVH test IS the culling. It would be more about unloading/simplifying objects thar are far away or completely hidden not necesarely by camera angle, but by overall reachability to light from a more wholistic perspective (like culling away the stuff inside a completely enclosed building)
Last edited:
Similar threads
- Deleted member 2197
- Games
