Is it normal that for small element size value Fractal Noise COP generates a visible grid pattern?

Yes, I am seeing the same. Increasing the Tile Size should make "better". I do not think this is Blue Noise so there will be tiling...

https://blog.demofox.org/2018/08/07/tiled-blue-noise/ [blog.demofox.org]

Its just the nature of the beast with fractal noise I think.

Thanks guys. Yes, it seems that 4d simplex noise of Torus pattern type doesn't like small element sizes. I can't increase the element size, so I ended up breaking up the grid pattern by multiplying this noise with another one. It worked well.

Still, out of curiosity, I'd love to know why exactly this happens. Do other types of noise also exhibit this behavior?

This is actually something that I am researching as a masters student and it is a very interesting and worth while pursuit. As higher-order sampling and masking patterns are everywhere in physics, rendering, and compression.

It comes down to the tile size and the randomness it exhibits. True random noise is often expensive to generate. For example Nvidia research renderer Falcor uses precomputed patterns: https://github.com/NVIDIAGameWorks/Falcor/tree/master/data/bluenoise [github.com]

Some articles and resources that are inspiring in this aspect:

https://blog.demofox.org/2017/10/31/animating-noise-for-integration-over-time/ [blog.demofox.org]

http://www.geometry.caltech.edu/BlueNoise/ [www.geometry.caltech.edu]

Optimize the Fourier Spectrum of White Noise into Blue Noise:
https://bartwronski.com/2020/04/26/optimizing-blue-noise-dithering-backpropagation-through-fourier-transform-and-sorting/ [bartwronski.com]


VIDEOS too: