Can confirm, probably makes sense to allow for zero iterations...

T68076 can be tackled hand in hand (might be related even)

Can confirm, checking...

In T68944#763645, @Stefan Werner (swerner) wrote:

@Lauri Hastholm (laurijh) I don't have a CPU without SSE4.1, it would be great if you could leave feedback here if my fix worked for you. On unsupported CPUs, you should now see a message in the node itself and it should now act as a passthrough instead of returning a black image.

Commit d547f9d3d291 breaks builds, it's missing the BLi_system.h header on source/blender/editors/space_node/drawnode.c.

@Lauri Hastholm (laurijh) I don't have a CPU without SSE4.1, it would be great if you could leave feedback here if my fix worked for you. On unsupported CPUs, you should now see a message in the node itself and it should now act as a passthrough instead of returning a black image.

I've mentioned this requirement in the release notes now. But I still think there should be some indication in the user interface.

@Jeroen Bakker (jbakker) Maybe you could have a look?

Ok, thanks @Marcus Papathoma (machieb) that's the explanation. On Linux /proc/cpuinfo shows the CPU feature flags which for me includes sse and sse2 but no higher.
Us with an old CPU but a newer Nvidia GPU can still use the Optix/Cuda denoiser and related Blender add-ons.

I think for AI Denoiser you need a cpu with SSE 4.1 or greater. Your cpu is too old!

AMD Athlon(tm) 64 X2 Dual Core Processor 4400+ (family: 0xf, model: 0x6b, stepping: 0x1)
cpu MHz: 2300
cache size: 512kB (x2)

I can't reproduce the problem. Which CPU model do you have?

Having the same problem with OIDN giving a black render result every time.
blender-2.81-daba3e871ff5-linux-glibc217-x86_64
Linux 4.15, 8GB DDR2, GeForce GTX 970, Nvidia driver 430.26

Glad I could help! To answer your other question, the node setup you have is definitely not the best way to save render passes (although I've been forced to do the same thing in order to work with legacy pipelines). Look into the Multilayer OpenEXR format. Choosing this option in your main output settings will cause all enabled render passes to automatically save to a single .EXR file without any node work or technically incorrect color transforms.

Wow, I didn't even know you could mute node or how I would have done that. Thank you so much ~

The appended node groups in your example file are quite simple so those probably aren't the cause of your issue.

here work fine with @Evgeny (mortuusars) .blend test file on linux Blender 2.81 daba3e871ff5 (2019-08-20) AMD FX8350, 16GB DDR3, GTX690 2+2 GB VRam- Nvidia drivers 435.17, Factory Setting applied, LoadUI and Trusted Source disabled in file browser

Same here.
version: 2.81 (sub 3), branch: master, commit date: 2019-08-24 07:02, hash: b7f786c0b150, type: Release
GeForce GTX 750 Ti OC
AMD Phenom II X4 965 CPU
Win 10 x64

Sounds good to me.

The compositor is running separate threads for independent nodes. In most constellations, that means that all denoising nodes run at the same time.

There is a bug report regarding memory usage here: T69006: OIDN: RAM increases proportionally with each Denoise node used.

I still have to read through all of the above mentioned tests/nodesetups, but do we agree here, that this is not really to be considered a bug?

I'm not sure if and how this is handled by other software using Cryptomatte. Presumably there would be some metadata on the OpenEXR pass, or the names of channels are different than RGBA indicating that it is non-color data? We should try to match whatever the standard is if possible.

In T68239#758664, @Jacques Lucke (JacquesLucke) wrote:

If I understand it correctly, the cryptomatte passes should never be modified by some color space transformation. Is that correct?

Any eta. when this denoiser will be in daily builds that are aviable on blender site? (linux)

I think it's potentially confusing that we have a "Denoise" Cycles feature configurable through View Layer settings and an identically named "Denoise" node in the Compositor which is a completely different denoising system.

any news on this would be very usefull.

The windows libs have been added to SVN.

It can land now.

@Brecht Van Lommel (brecht) should I land the patch in its current state (OIDN disabled by default) or do we want to wait for install_deps.sh to be fully working first?

In D4304#124368, @YAFU (YAFU) wrote:

Hi.
Will the compilation and installation of OIDN dependencies through install_deps.sh be automated?

Turn off OpenImageDenoise for full/release builds until we have libraries.

Another option would be to only have a single horizontal extension stored that is used by all 4 curves (limits current implementation, but perhaps easier to understand). The second option will break current files.

Hi.
Will the compilation and installation of OIDN dependencies through install_deps.sh be automated?