Discussion
Ray-tracing is a technique combining fast hardware with clever software to render photorealistic images! Most movies today, and increasingly games, are rendered using ray-tracing.
To be clear: Ray-tracing isn't really a new thing, its just getting more popular now that our hardware can afford it. Some of the earliest CGI was raytraced (using Signed Distance Fields to model their geometries), & build-time "baking" raytraced shadows into scenes has long been common. Myst was raytraced!
1/?
Ray-tracing works by simulating the paths of numerous random light-rays from simulated lights to a simulated camera. Or rather we simulate in reverse (formulas are the same) so significantly more of the lightrays are relevant. Though this does introduce biases to correct.
We arguably never will be able to simulate every single light-ray, but we can simulate enough.
To efficiently decide which surface a light-ray bounces off of we use data-structures called Bounded Volume Hierarchies.
2/3?
A BVH splits a 3D model into a tree of bounding-boxes, a self-explanatory name, so we can rapidly check a few boxes along the appropriate branches of the tree rather than checking than checking every triangle in that model. Thus minimizing how much work needs to be done per bounce.
From Sebastian Lague's experience implementing this I suspect the real trick to hardware-optimized ray-tracing is to provide more opportunities to optimize BVH lookups.
3/4!
The final trick is to bias the random selection of light-rays towards those which contribute more, & (again) correct for that bias.
There's modern tricks for cheaply generating suitably-weighted random numbers.
So what's cheaper: The traditional OpenGL pipeline or ray-tracing? More than most things in compsci: It depends.
How fancy do you want to your render? Ray-tracing has a higher baseline cost but you won't pay more for shadows, reflections, refractions, portals, blackholes, etc.
4/5!!
Install bonfire.cafe
Get the full app experience
