Realistic atmosphere

[msvcrt (rrk)]

Hey TheJamsh!
Thank you for purchasing!

@TheJamsh;304299

I also only just found out that Custom HLSL nodes can be made to talk to and call each other from within the same material graph, which is quite cool.

I was considering using the Custom Node…
Simple code implementation of loops would be much cleaner than those long chains of sample nodes…
However, Custom Nodes are platform dependent. Besides, the process of editing code for the CN is currently awful.
So, I decided that implementing everything with regular nodes is a more reliable way.

@TheJamsh;304299

Just checked it out. It’s quite performance intensive, which I kind of expected to be fair.

Unfortunately, yes. But I’m afraid this is as much as I can get from the realtime implementation. My first implementation had about twice greater instruction count and I spent some time reducing it as much as possible. Also, I created different material instances with different sets of enabled features. Take a look at the MaterialInstances folder’s content: R = Rayleigh enabled, M = Mie enabled, S = Volumetric Shadow enabled, T = Translucency enabled (skies only). If you want further improve performance, you can decrease number of samples taken in AtmoSampleDensity and AtmoSampleScatter functions, but quality can suffer a bit.

@TheJamsh;304299

What’s weird though is that the performance seems to climb up and down even if I’m sitting in the same spot in a scene. Haven’t tried it with a totally blank project so I’ll give it a go. Just wondering what kind of performance you get with this? I’m on a GTX 980 with i7.

I’ve tested performance on several machines and even got good performance on a laptop with i7 + GT 650 (but with 1366x768 resolution). Never noticed performance changes with static camera. Performance mainly depends on how many pixels on the screen are covered by the material, as the pixel shader is especially computationally-heavy.