Just stumbled upon this ancient PcPersective article on real-time ray tracing.. In 2008. :) pcper.com/2008/06/ray-... It details the story of nine students (Karim, Wussie, Rick, Jan, Rutger, Trevor, Mathijs, Roel, Wilco) who worked in various roles on CPU-ray traced games using the Arauna engine.

Some images showing off my improvements to my photon mapper I have integrated into truetrace The photon mapper is up on the Dev branch on the github

Forgot to post these From some improvements to my caustic photon mapper tacked into truetrace last week

Anpther major update to our cluster lod ray tracing sample. Zorah scene available as glTF raw.githubusercontent.com/nvpro-sample... BLAS caching github.com/nvpro-sample... BLAS merging github.com/nvpro-sample... Special thanks to Adam Marrs, Pyarelal Knowles for ideas and Chris Perrella for zorah.

TinyBVH 1.6.7 is now available on the main branch. Changes: * TinyBVH now only needs C++11 (tiny_ocl.h needs C++17) * Full-sweep SAH builder is now threaded * Fix for crash in BuildAVX Still zero dependencies. :) And: The repo reached 1k stars! Link: github.com/jbikker/tiny...

I am glad to attend #HPG2025 where I just presented our paper "Fused Collapsing for Wide BVH Construction", co-authored with Mathias Paulin. We propose a fast build algorithm for wide BVHs that directly computes a wide hierarchy without additional collapsing pass. Webpage: wbrbr.org/publications...

Rendering nerds! Check out our latest work "Vector-Valued Monte Carlo Integration Using Ratio Control Variates" that has just gotten the best paper award at SIGGRAPH 2025. This paper presents a method that reduces variance of a wide range of rendering and diff. rendering tasks with negligible cost.

A Fluorescent Material Model for Non-Spectral Editing & Rendering Belcour Laurent, Fichet Alban, Barla Pascal (Intel Corp. France, Inria Bordeaux) arxiv.org/abs/2505.19672

O wow, ReSTIR keeps giving. Awesome stuff.

Saw that "The Witcher 4" UE5 tech demo with the voxel trees for fast rendering at a distance? I figured... that should be easy to do in TinyBVH. :) So now TinyBVH can do voxel meshes. Attached video: CPU-only, switching between BVH and voxels. Code in tiny_bvh_foliage.cpp: github.com/jbikker/tiny...

Now available: A blog post on (S)BVH optimization. The article describes how to improve beyond SBVH quality, often significantly. Results are compared to full-sweep, binned SAH and the H-PLOC. In short: Off-line BVH construction can double ray tracing performance. jacco.ompf2.com/2025/05/20/b...

Added tri presplitting thanks to BoyBayKiller Before vs after tri presplitting(splitting triangles before doing ANY building, vs SBVH which splits DURING build), only affects SWRT Tri count jumped from 2832120 triangles to 3348607 triangles

Here is a new blog post of mine, where I dive deep into the theory and math behind ReSTIR GI. Big thanks to Markus Kettunen for verifying the theory, reviewing the article, and catching a key detail I had missed. agraphicsguynotes.com/posts/unders...

That's great news! Very interesting work.

Progress on the quest to the Ultimate BVH. Two observations: 1. You can optimize a BVH, using "Fast Insertion-Based Optimization". 2. "Spatial Splits" (SBVH) beats an optimized BVH, narrowly. Two challenges: 1. SBVH practically must use binning 2. SBVH doesn't benefit from optimization. Now what?

pathtraced Audio vis thing, but as usually I dont have enough ideas to take it very far ASVGF is doing a bit of smear, but honestly not as much as I woulda thought

And another one from Yanus playin with the photon mapper I’m playing with in truetrace

Behold! First speed! Some experiments from Yanus with the photon mapper I am playing with in truetrace

Grading for Y1/Block C @ Breda University is complete. Our students used path tracing, BVHs, reprojection, blue noise, post-processing, importance sampling, MIP-mapping, bilinear interpolation and tons of other ingredients to build a CPU ray tracer + game. Behold, the gamedevs of 2028!

Baby’s first photon mapper!

1 sample per pixel and 2.6k samples per pixel(cuz glass takes awhile)

well somethings happened No denoiser except for occasionally TAA Just raw restir for the most part

so now that I can disable rasterization entirely in truetrace... After turning on RT cores, this is the perf I get(and also turning off my postprocessing, cuz its a bit expensive) (With RT cores off, perf bounces between 60 and 110, average of 75-80)

A bit better This time is WITHOUT RT CORES, as I found why it was slowing down before(Always merge your BLAS's people!) Still 195 dynamic objects, 5 mil tris FULLY pathtraced using my renderer TrueTrace

So I am not quite sure why its so low res/artifacty, I need to look into the recorder I use again I guess But here, another high speed test I AM using RT cores, but without RT cores the fps is about 55-60(still not enough...)

Too many samples... the first image took 20k cuz my volume sampling is still very inefficient... First image was mesh lights, second was point lights its clear that I need to find a way to better sample both the SGTree and the RIS for unity lights along a finite path...

Trying the classic Unity Courtyard, but NO baking 100% FULLY pathtraced with RT cores About half the cost per frame is not even truetrace Ghost artifacts were some missing meshes Truetrace is completely free and open source on the asset store and on github for more up-to-date versions

I wanted to compare two images here Left is naive PT The ONLY thing thats changed in the right image is I turned on my restir GI. thats IT! and not including the extra shadow ray, its only 2ms to run... #pathtracing #unity

Oh also this Really need to improve my volume sampler as this took 40k samples purely cuz of the volumetrics... 100% pathtraced in unity

Quick experiment with diffuse transmission Right side is with diffuse transmission on, left side is with it forced off Scene is from sketchfab, which I will put the link to in the replies here Fully pathtraced

0.976% of the paths traces per frame, vs the image rendered by those paths

Was massively optimizing terrain(0.9ms for a single full trace down to 0.3ms), and improving its quality(far fewer holes, and handles far distances much much better) And I ended up In the train scene, and look at those reflections!(ASVGF denoising) No RT cores ofc #pathtracing

More fog experiments

God I wish I had a larger map for these - I just REALLY love trains Trains are from WSM Game studios(link in replies) But anyways, I was testing moving objects and shadow response for ASVGF ALSO, the asset store for TrueTrace has been updated!(link also in replies, crossing my fingers its fine)

Using ReSTIR to importance sample the brightest pixels on an image www.shadertoy.com/view/XfycWw

And a daytime one

Look at the shadows! ASVGF is way more stable now!

Just a displacement texture acting like its a cutout texture

So I did a big optimization for the GPU light BVH builder 20k triangles is down from 6ms down to 4.8ms average Unfortunately that is difficult to show, so here it is in action, NEE light sampling ONLY; NO NAIVE LIGHT SAMPLING OR MIS!

Been implementing this paper: arxiv.org/pdf/2502.02011 and its awesome!

Truetraces new maximum triangle count is 41,297,762 triangles(Before instancing)

STOP USING GOOGLE. If you keep using their service, they still get paid. If the revenue keeps coming in, shareholders stay happy.

Progress comming along nicely with the gpu bvh builder I really wanna figure out how to do the cwbvh on the gpu too

I quickly added a "time" node. Now I can animate!