Ray reuse! Woo! MUCH faster convergence for the GI now. While still being under 2ms per frame. GI also looks a lot less grainy. At this point, I mostly just need to make the probe placement better. I'm probably gonna use the SDF to help with that.
Ray reuse! Woo! MUCH faster convergence for the GI now. While still being under 2ms per frame. GI also looks a lot less grainy. At this point, I mostly just need to make the probe placement better. I'm probably gonna use the SDF to help with that.
0 vue
Publié le 05/11/2026 15:51
1:35
Ray reuse! Woo! MUCH faster convergence for the GI now. While still being under 2ms per frame. GI also looks a lot less grainy. At this point, I mostly just need to make the probe placement better. I'm probably gonna use the SDF to help with that.
2 commentaires
?
It also means you need very few rays per probe to actually achieve the effect. So instead of like 96+ rays per probe you can feasibly get away with 16-32 for your primary traces. This also heavily uses importance sampling to steer rays closer to light sources, so this is _heavily_ biased.
The basic idea is that each probe's clipmap (think of it as "detail levels" in a sense) feeds into each other's clipmap with a sort of ray history that can be reused. The result is ray count matters less and scenes that need eventual convergence converge a lot faster with much smoother results.
Bending light again, spacebirb?