Color Noise Reduction vs Splitter Luma NR

I hope this is not an old question...
The question basically is: how does Chroma NR work in Davinci?
I mean, Spatial Threshold Chroma should handle color noise, and it does the work pretty well, but how does it operate under-the-hood? Does it split the image into RGB in order to apply 3 different luma NR amounts to the 3 channels?
I'm wondering this, because recently I tried a different approach: instead of using the "classic" Spatial Threshold Chroma on a single node, I used a splitter/combiner node to separate RGB, and then I applied 3 different luma-only amounts to the 3 different channels. It seems to me (but I might be wrong) that the result is better (because every channel has a different amount of noise, and once saparated the channels, the noise is presented as BW-only noise).

Please, let me know if you think this is only an arbitrary impression, or if you think I'm in the right direction.

Thanks.

I am guessing it will internally convert to YUV space, but it might keep it in YRGB space, which is DaVinci's "secret sauce" colour space where chroma is kept as separate RGB channels instead of a G-R and G-B difference channel, but it would probably be slower so for performance reasons I'm betting it will internally process images in YUV, where "chroma" is UV. You can read about it on Wikipedia.

If you want different denoising amounts for each YRGB channel (you might want to if for instance due to extreme white balance one channel is pushed hotter than the other channels like the blue channel under incandescent lighting), you can split them off using the primaries bars and merge them together with a layer mixer set to additive blending:

you can split them off using the primaries bars...

Wow... never tried yet... I'll surely try!

Thanks a lot!

You can also decouple Chroma & Luma in both TNR and SNR. One recommendation I've seen several times in the past is to use TNR for Luma (Y) only, then use SNR for Chroma (C) only. That way, the SNR does the least amount of damage to apparent sharpness.

In some cases where I have a lot of noise mainly in one channel, I'll use a piece of a splitter node and use the NR only for Blue (as one example). For this, you need to work in Luma-only mode for either TNR or SNR.

Decoupling with the better modes only works in version 15.

Many thanks to both of you!
Great advices!

Just to know: if I overwrite my old Davinci 14 folder with the new 15, I should maintain all my old projects, right?

I would always back up first. DR15 is still in beta!
Plus, once you start DR15, it'll ask you to update your database and after that there is no way back.

once you start DR15, it'll ask you to update your database and after that there is no way back.

Well... I'll stay with 14 until 15 will be out of beta testing.
Anyway, just to know, I've seen something in v.15 about the metadata for HDR10. Does it mean that for HDR videos we don't need Hybrid/FFmpeg maxcll/fall more?