Cody Bennett

@Ademola_4life Some helpful links here. x.com/CodyBennett/st…

@CodyBennett Amazing! Im curious what the octahedral mapping decoder looks like

Have been optimizing to run on Apple TBDR and mobile. No hardware raytracing but custom compressed binary BVH with TinyBVH for coherent rays. Timings are per-frame to compute a single bounce from scratch, no temporal accumulation. Apple M2 baseline 1x / 2x (supersampled)

@Ademola_4life Yes, and some more technical details. x.com/CodyBennett/st…

@Ademola_4life x.com/CodyBennett/st…

@garrettkjohnson So, I think it's useful to have a builder discern whether it's being used for coherent rays (primary view, physics/gameplay) or incoherent rays (secondary, stochastic/path-tracing). A combination of 4-way and CWBVH should do well. Would need to further research for TLAS also.

@garrettkjohnson For reference, it would look something like this. Identical to 2D with direction-major storage, except using octahedral-encoded rays and depth-aware upscaling for the extra z component. I also do a 1px trace during composite so it resolves to 1px accuracy. shadertoy.com/view/ffl3Rf

@MrCollison @majidmanzarpour @threejs @garrettkjohnson Amazon Lumberyard Bistro is a good one I think.

@CodyBennett @majidmanzarpour @threejs @garrettkjohnson My intuition exactly. Also, I wondered what it meant when I saw "external example"! I will find a way to publish it, portability is gonna be the biggest challenge so just example scenes to start with. Do you know of any big outdoor+indoor scenes to stress test with? Even paid?

idTech 8 style Global Illumination (FAST AS HELL) in @threejs (via WebGPU) I've worked on this for 5 weeks (on and off) and it's still not there, but all the core tenets work. Next more denoising + better temporal accumulation, and some larger scale examples. (Running at 4K)

@SheriefFYI @minitbnn Decals, screen-space mesh blending, and anything evaluating a BRDF, e.g., SSR. All rely on primary visibility and naturally benefit from a gbuffer. Many world-space GI not only use some sort of screen-space AO for missing detail (foliage, high freq) but guiding or small traces.

@minitbnn what post effects need full g-buffers versus just color, depth, and motion vectors?

one of my hottest takes is that deferred rendering's benefits flip sharply once you cross a certain resolution threshold as bandwidth consumption begins to choke your performance and games should have explored things like clustered forward rendering as we approach 4K.

@MrCollison @threejs How do you mitigate leaking of irradiance volumes, or is it not much of an issue with DDGI-like visibility? Maybe something for Tiago but was a Surfels-like caching method or "Radiance Caching with On-Surface Caches for Real-Time Global Illumination" explored or left due to time?

@MrCollison @majidmanzarpour @threejs I don't think this would be tenable for Three.js to maintain as an example but rather as something external. If you publish it somehow, Three.js has incorporated external examples lately such as @garrettkjohnson's work. Or better, perhaps it could be showcased on the website.

@majidmanzarpour @threejs Thanks Majid! I'll open source it in due course - I wanna make sure it works across a range of scenarios so people can just plug + play I'd submit a PR to @threejs but it's raw wgsl in several passes + files to piece it together, so it may be too bloated for an official example

This scene shows both GI & raytraced reflections clearly. It's also great for debugging noise because it's fairly flat and uses only one directional light. This scene would be basically unlit without GI, yet look at the indirect light below the fridge & the counter reflections!

@dangreenheck I have implementations of their later work, and it's difficult since the presentations leave much out on lighting, and most online references are on Horizon: Zero Dawn, including the chapter. I think a physical basis like RDR2 or Unreal would go further. x.com/Cody_J_Bennett…

I wanted to share an update on Three.js Sky Pro development for those of you wondering why I haven't launched it yet. tl;dr: clouds are hard. For the past few weeks, I've been trying to replicate the results of the cloud systems developed by Andrew Schneider and team as Guerrilla Games for the Horizon: Zero Dawn games. This is basically the state of the art for real-time clouds. Unfortunately, I have been unable to replicate the results, despite poring over the papers, talk notes and book chapters on the topic. I can get close, but there's a lot of special sauce that isn't shared in the resources. This special sauce is what separates okay-looking clouds from "blow your socks off" clouds. I'm not going to sell something that I think sucks. My name and reputation is attached to it, and I have no interest in being known as the guy that made "okay" clouds. So I'm going to take everything I've learned and try building my own solution. I have no idea how long it will take. I have no idea if I'll be successful. But I'm excited to give this tough problem a go. Apologies if you're waiting on this, but I promise the end result will be worth the wait!

@garrettkjohnson github.com/nvpro-samples/…

@garrettkjohnson They are slower to trace, but faster to update is a big one. I have been slowly working on a conservative error metric for offscreen LOD selection too rather than just clipmaps and hoping shadow casters aren't prominent enough (distance from camera not conservative anymore!).

Unsurprisingly, keeping all meshes in a TLAS has a severe path tracer perf impact when there are lots of overlapping bounds - slower than the WebGL version with merged geometry in some cases! These images show the ray traversal cost from a 46 mesh TLAS vs pre-merged geometry.

@garrettkjohnson With Radiance Cascades, I found that a BVH2 is better than CWBVH for BLAS, as structured rays are directionally and spatially coherent. I don't see much on this online, as this would be roughly equivalent to optimizing BVH for primary visibility, which would compete with raster.

@garrettkjohnson Was just thinking of this after implementing CWBVH + H-PLOC. I precompute CWBVH to BLAS for my clustered ray tracer but don't stream them like I do for my rasterizer (two different DAGs). Awkwardly, I cannot specify up front for HWRT with a streaming scheme.

@MrCollison @kvickart @threejs This is not good. Unreal Engine is not permissively licensed, to the point where I will not look at its source code. DDGI is well understood and not complicated to implement, except maybe depth moments. Just do that or even look on to Godot's documentation on SDFGI based on DDGI.

@kvickart @threejs I'm half wondering if Fable could - if I also pointed it to the unreal source (not sure I'm even allowed to do that) But even reading the UE source code on GitHub and handholding Opus has been hours across 3 nights to even get this far

Implementing UE5-style Lumen lighting in @threejs ... It's going ok?? Clearly there's something very funky as the SDF reflection for the cone looks like it's suffering from axisitis Let's see where this one goes with a few more rounds of review & prompt 🫡