^^ I think the RTX Mode ON (10 ray per pixel crazy bounce of mirrors with a simple surface) shows how trivial the ray/triangle interection tests are on the RT core (it runs really well!) vs. how much more expensive shading of the results are with "RTX off" and less ray/pixel. The performance difference in context aligns with what we have been seeing across BFV and the Northlight presentations.
No, it should be much better. In BFV the shading cost is high, probably because they have many shaders so divergence, also they need to iterate over shadow maps to shade hit points.
Here there are no shaders used at all. Instead one random light is chosen and a shadow ray is necessary to determinate light visibility, plus a single texture fetch to get aldebo.
So, if 10 mirror reflections work with good performance, you could use 5 indirect bounces instead (one ray plus one shadow ray each). Performance should be very similar. (In a modern game with complex geometry perf would drop, but non the less this is really good)
However, i assume it would give best IQ to use just 2 bounces but 3 samples. Eventually 3 samples prevent the filter from washing important high frequency details like contact shadows.
In any case the game could look a lot better with a bit of tuning.
This is one I am thinking about. Isn't the usual concept of texture filtering not at all the same here with this path traced q2 since everything is path traced? Isn't the textured surface and its entire look reconstructed/composed during the denoise step? That is its "filtering", right?
No. The filter works only on incoming light. The textures are composited later with this result similar to deferred shading (see earlier posts).
So the only spot where textures affect the filter is their contribution to indirect lighting, which still gives a smooth signal on the filtered result.
(see src/refresh/vkpt/textures.c (579))
