Rasmus Rønn Nielsen

Looks great! Doing anything fancy for denoising?

www.phys.ufl.edu/courses/phy3...

I see now that I accidentally swapped DXR1.0 and DXR1.1 in my previous message. Sorry for the confusion, I hope the message came across.

I thought the problem of variable material complexity was handled in DXR1.0 by assigning different hit shaders for each material? 🤔

Explore GPU advancements in M3 and A17 Pro - Tech Talks - Videos - Apple Developer Learn how Dynamic Caching, the next-generation shader core, hardware-accelerated ray tracing, and hardware-accelerated mesh shading of...

Like AMD, Apple too bet on a DXR1.0-like hardware implementation of ray tracing (as opposed to Nvidia/Intel which is more DXR1.1-like) and I guess this is one of the primary reasons for why they both have prioritized dynamic register allocation. I wonder how these two strategies will play out.

Coincidentally, I’m currently working on a little test suite which tries to measure the performance of various temporal filters (Stachowiaks being one of them). Given a stream of samples from a dynamic distribution, which filter “finds” the current mean the fastest without being too noisy?

Yeah I agree it lacks detail. You can find a little more detail in media.contentapi.ea.com/content/dam/..., see page 26. Personally, I want to understand more deeply why+how their method works and if we can do even better.

Nice! For the blending, did you go with Stachowiaks method from Pica Pica or something else?

The "middle ground" that I suspect deserves more attention is realtime GI systems which solves in a coarser space than screen space. GIBS is one example (surfels), Godot has another (DDGI-like probe cascades), but can you name more? Potential benefits are no SS artifacts and fewer rays to trace.

Agreed. Other "middle ground" solutions are definitely possible but this solution space is largely left unexplored (perhaps because this is less sexy?). IMO this is a shame and FWIW we are exploring this space at Unity currently. Also, you may enjoy this prototype of mine: x.com/raroni86/sta....

Interesting as always, Tomasz. I have a question! ✋

Do I understand it correctly that each full-res pixel will only contain contribution from 2 lights in the end, i.e. the two that were output from WRS? Or do you also use the 6 "losing lights" somehow?

Great work, Jacco, and thanks for sharing. Do you plan to make a GPU-side CWBVH builder?