Next gen lighting technologies - voxelised, traced, and everything else *spawn*

You forgot the Medium DXR settings in BF5 where GTX 1080 Ti about matches RTX 2070
Perfect mirror reflections (Medium preset in BF V) work fast because rays are coherent, hence a SIMD can process a bunch of rays at the same time (for Frustum-Traced hard shadows, it was something like 5*10^9 rays/sec on TitanXP), with higher material roughness, rays will diverge and follow different paths quickly, this hits SW RT performance badly (performance in Metro Exodus, where all rays are incoherent, is a good indicator of this).
 
Last edited:
BTW, tried SEUS minecraft mod today and it features screen-space GI only (which is also explicitly said in the mod's settings)
I haven't noticed any screen-space GI caching in textures, voxels, or in other resources, it behaves exactly as a simple screen-space method - short GI radius, disappears at screen corners, etc.
So... no wonders here
 
My bad, I just figured that the latest SEUS PTGI E6 release doesn't work properly with 1.13 game version, so world-space GI was broken.
Just tried with 1.12.2 and now it works, there are tons of bugs and other weird issues with lighting in enclosed spaces, probably something is still broken in my build.
 
I think the real differentiator will be area lights firstly

On the topic of area lights ... since games have already implemented irregular Z-buffering, I'd like to see this method for area lights in games. It's a brilliant way to do almost exact shadows for a large number of point lights approximation of an area light and just like irregular Z-buffering I think hardware has caught up to the point where it has started to make sense (maybe not with the 128 point light approximation they used, you could use 32 but vary their position per pixel for a bit of dither).

I don't think raytracing will be able to compete with this, even with the lack of on the fly BVH occlusion culling for rasterization (purely an API issue, not fundamental).
 
Last edited:
I'd like to see this method for area lights in games.
Nice paper.
We could say 'irregular Z-buffer' is just a buzzword for binning to a regular grid. Again you have to deal with compaction of huge amounts of data, similar to the RT problems i've mentioned earlier. This is also a bottleneck in many other things, e.g. SPH fluids and collision detection in general.
GPUs already are quite capable of doing this very fast, but if it could be done faster with FF this would be a very welcome generic feature - together with fine grained device side dispatch ofc.
I still hope upcoming AMD hardware goes more into this direction than into 'RT cores'. One could still call it hardware accelerated RT then, so recent announcement 'PS5 has RT' still tells me nothing about the future... damn :D
 
On the topic of area lights ... since games have already implemented irregular Z-buffering, I'd like to see this method for area lights in games. It's a brilliant way to do almost exact shadows for a large number of point lights approximation of an area light and just like irregular Z-buffering I think hardware has caught up to the point where it has started to make sense (maybe not with the 128 point light approximation they used, you could use 32 but vary their position per pixel for a bit of dither).

I don't think raytracing will be able to compete with this, even with the lack of on the fly BVH occlusion culling for rasterization (purely an API issue, not fundamental).

Just glancing at that paper it appears that the performance of the algorithm completely collapses for any reasonably large triangle count. Going from 30 FPS on very low poly scenes (only a couple thousand triangles) to less than 1 FPS for medium poly scenes (a few hundred thousand triangles). I realize it was running on old hardware, but modern game scenes can be expected to have a few million triangles... And render more than just shadows...
 
Two "1080Ti"s in SLI (hammered to a 100% utilization each), couldn't approach half the performance of a single 2080Ti in most scenes with RTX on in Shadow Of Tomb Raider.

Since on topic of SLI; can they do SLI for RTX cards doing RT?
 
Wasn't sure where to post this, interesting though. Using Re-shade with a new screen-spaced path traced global illumination.

But can it run Crysis... path-traced?

 
Wasn't sure where to post this, interesting though. Using Re-shade with a new screen-spaced path traced global illumination.

But can it run Crysis... path-traced?

SSGI methods certainly will become more common in future, hopefully many of them will be backed up by proper scene data like in Metro.

.. Actually even standard perspective corrected cubemaps and such might be enough to allow SSGI get a lot better handling of environment. (Miss or invalid data from SS and sample cubemap with same ray.)
 
Last edited:
Back
Top