Listening room

[Jeremy Hill]

Here’s a scene put together for no real reason, which seems to be asking me to make it into something evocative of a Kubrick shot.

It uses some models from grabcad and flyingarchitecture, as well as textures from cc0textures. Not having the luxury to spend too much time with modeling, I really appreciate these resources, and encourage you to pop over, check them out, and maybe show your support.

[Philip]

Very nice!

[Eric]

Is there an efficient way in Bella to add microbumps for polished floors and rainy pavements?, When I try with a fine tiled grid of bumps my Maxwell renders are slower, I think it has to do with the moire patterns caused by the grid although I don’t see any moire in the rendered pixels they must be visible if I zoomed in. Next time I will build a more organic larger-area bump map and check if that’s quicker.

[Jeremy Hill]

Bump mapping has an associated cost, since you must evaluate more than one point to find the curve, but this cost is not related to the nature of the map, or really anything else. It will increase as the visible bumped surface area increases, as a percentage of all rays being computed (e.g. as you zoom in on a bumped surface).

So I guess if you are able to observe a measurable difference unrelated to that, it could be related to the lighting situation being altered by the bump; it seems possible that data will be forced out of cache more often than if you did not use bump, since bump will increase ray incoherence.

[Eric]

This felt like a 30-50% slowdown, versus no bump.

This is the texture I used with very high tiling    I will try again with a larger and more random map.

[Jeremy Hill]

For what it’s worth, normal mapping may have a slightly lower cost than bump mapping (in bella, not sure about mw — the whole texture system works very differently in bella), but I have not yet specifically measured it.

[Eric]

A micro bump effect built into a shader would surely be the quicker, but the least flexible too. For random rain effect bumps, might as well use a big map.

[Jeremy Hill]

Something like the texture you posted, which just looks like U & V sine waves interfering, would be a natural candidate for a procedural texture. I’ll put that on the to-do list. As far as such things being “built into” bella materials, the way they work, this is really just another way of saying procedural texture, since to get the bump for a given ray impact, you are going to be evaluating the UV coord at the impact point to find out how you need to bend the normal.

[Eric]

I might run some torture tests to bring out the moire patterns, so as to find the most tolerant high frequency pattern.

 

If we had a procedural random bump with no repeating pattern, that would be moire free. Like this tile but infinite;

[Eric]

Ran a proper speed test with the ‘sine waves’ normal map, vs no bump, and there really is no slow down, it just takes more sl to resolve the floor.

Will try and make a larger-area random microbumps map next.

[Jeremy Hill]

When you say micro, I’m curious about the scale you mean — for example, I assume you mean you are rendering with repeat values that make the period of each sine smaller than a pixel in the output image, but (if so) how much smaller?

[Eric]

I don’t know which is the optimum size, The micro bump repeats at 1000, and the diffuse concrete which is about 2m realworld is set to repeat at 0.1, so overall 10,000 ‘sine-wave’ per 1 diffuse.

[Author]

Posts