Patrons Ask GN 39: Ray vs. Path Tracing (ft. Andrew), Undercover Snowflake, Zen 3

Used to work with some other poeple on an indie/open source game project, in a past life :)

Max Eliaser

That is a really great post - much appreciated. Do you work in the industry?

Hi GN team; I would like to hear more about the machine learning and technical aspects of RTX/DLSS and everything that AMD is putting into their cards. I feel technically incompetent when I hear these terminologies and it would be great to introduce some basics about how these technologies come together in the background (think computer science lectures).

AMD raytracing = RayMD ofcourse. 😁

Azdeus

If you're interested in less-demanding alternatives to full raytracing, check out radiosity as well. Radiosity is a good technique when your scene is mostly static and you want to support a large number of light sources rather than just one or two. In radiosity, the number of rays that need to be simulated is equal to the square of the number of surfaces, rather than being proportional to the number of pixels or anything crazy like that. Rather than bouncing rays from the camera until they reach a light source, radiosity starts with a list of all surfaces and computes how much each surface illuminates each other surface (if they are able to illuminate each other at all.) If two surfaces can be connected by a ray without hitting obstacles, then you figure out based on distance and relative angle what percentage of the light from surface "A" will transfer onto surface "B." Once you're done building your list of percentages (coefficients) you can reuse them to compute as many bounces as you want. You do direct lighting in a traditional non-raytraced way, and then use your list of connected surfaces and bounce coefficients to compute the indirect light. The lighting on the surface is thus the sum of the direct lighting with a number of bounces of indirect lighting (four to six bounces is usually pretty good.) To make it more fine-grained, you can slice up all the surfaces into grid squares to increase the number of surfaces and the number of rays that are traced. Baked radiosity was used in Quake II and realtime radiosity has been used in some more recent titles such as Battlefield 3. The benefit of radiosity is that you only have to do all your ray-plane intersections once, in advance, and then you can reuse them. This obviously doesn't work well for scenes where the geometry is moving around all the time because that would invalidate the precomputed intersections. It's also less precise than raytracing and cannot create effects such as a mirror reflecting an image.

Max Eliaser

Needs to be the next GN shirt

An Andrew appears from the wild! I love seeing on Andrew's expression how much is flying through his head trying to answer the ray tracing questions in terms mere mortals can understand. It's like watching ray tracing in action! Good stuff gents!

Mark_deLight