OIDN and Arnold in R21

[nwoolridge]

Not a question, but just a note: if you are interested in trying the new Intel Open Denoiser in C4D R21 with Arnold, be aware the Arnold's default filtering (gaussian at 2 pixels) prevents the denoiser from working. Change the filter to 1 pixel, or to one of the other options that doesn't have a pixel parameter, like box. Then it works extraordinarily well in my testing.

[lee_griggs]

Thanks Nicholas! Would you be so kind as to share your results here? How do you find it compared to Optix and the Arnold Denoiser?

[nwoolridge]

@Lee Griggs

Sorry for the delay, I've been squeezing these out on a 4-core MacBook Pro (2017) in amongst some other work. So, these were not done on a render power house, by any means...

The OIDN has very few settings; just the option to use multi-passes to improve the result. This means that, if normal and albedo passes are done, it can better detect object edges and texture detail, giving a better result. Overall verdict: very fast (1-3 seconds per frame), awesome for stills, potentially useful for animation, but finding the right sample settings for animation action will be challenging. My low samples are ridiculously low, but show the ability to clean up the quickest test renders.

The animations also show the potential, but also the flickering and smearing at lower quality settings (animations are at the same low settings: 2,1,1,0,1,0). None of these renders have the multiples option enabled, so results could be better if I fed the denoiser those passes. More extensive tests would look at fine surface textures and details, which can be smeared by the denoiser, and refraction, etc.

Animations:

Animation orbit

Animation wobble

[nwoolridge]

As always, download the dropbox links so as not to contend with the godawful DropBox recompression.