Macbook M1 max pro and noise reduction

Hey there!

I got the Mac M1 max pro with 10 CPU, 32 GPU and 64GB RAM. So it's the most powerful version of the M1.
Everything is well, but the noise reduction is the beast!

If I set it to best settings (5 frames, better) and give it about 30% (+ 30% chroma). I get 1-2 frames in quarter resolution.
It would be no problem, for I just can activate it and the end of color grading in order to export it. BUT while exporting its EXTREMLY slow, just in the speed of 1-2 frames.

Is it normal (for such 'best result' noise reduction) is are there some false settings or something else I missed?

Cheers
Alex

noise reduction is one of the most demanding tasks especially for video. Unless there some Ai breakthrough soon (that looks like Adobe LR denoise in speed and quality) currently a couple of high end Nvidia cards will be faster.

sure, it's a piece of hard work.
As it seems M1 Mac still doesn't have support for eGPUs, isn't it?

...I got the Mac M1 max pro with 10 CPU, 32 GPU and 64GB RAM. So it's the most powerful version of the M1...while exporting its EXTREMLY slow, just in the speed of 1-2 frames...

What codec? We need to know the resolution, frame rate, bit depth and chroma subsampling. Is it UHD 4k 10-bit 4:2:2 at 23.98 fps or what? If you aren't sure, you can use Media Info or other tools to verify this.

Video noise reduction (whether Resolve or Neat Video) is a multi-frame effect. That means it examines a sliding window of frames. There may be cases where if NR is applied as the last effect in a stack or node tree, it can force the recomputation of all effects before it -- for each frame in the window. IOW if the window is 5 frames, it may cause all effects before the NR effect to be recomputed 5 times for each frame in the timeline.

For that reason it's usually a good idea to apply NR as the first node in the tree. So procedurally, you defer NR until one of the last post-production steps, but when you apply it, put that on the first node. I think this is the right procedure, if not, someone will correct me.

What codec? We need to know the resolution, frame rate, bit depth and chroma subsampling. Is it UHD 4k 10-bit 4:2:2 at 23.98 fps or what?

The project has several files from different cams.
UHD 4k 10-bit 4:2:2 at 24 fps from Sony A7iv (trimmed from 60fps)
The others are 8bit and I think 4:0:0 (Sony A77, some Canon and GoPro Hero 5)

Yes, I applied it to the first node of the whole timeline. I think it's just too heavy for the Mac.

....UHD 4k 10-bit 4:2:2 at 24 fps from Sony A7iv (trimmed from 60fps)
The others are 8bit and I think 4:0:0 (Sony A77, some Canon and GoPro Hero 5)...

When you say 24 fps trimmed from 60 fps, do you mean the clips are actually 60 and have been rate-conformed to 24 fps? IOW is the on-disk clip 60 fps? What actual frame rate are the other clips?

60 fps H264 high L5.2 4:2:2
50 fps AVCHD
30 fps H264 high L5.0
25 fps H264 Baseline 5.0

Project is in UHD (export in HD) 24 fps

IOW is the on-disk clip 60 fps?

what do you mean by that?

yeah that sounds about right.

The Apple Silicon GPUs are currently not that fast when it comes to some heavy things like denoising.

In general - node/smart caching when using NR (either built in or from a plugin like Neat Video) is a good idea. It helps to have it done at the beginning of your node tree so that you can then get good performance downstream from the cached NR node.

You can render with the cache to speed things up (just don't use too much compression on the cache format). But if you need to render source res it will still be a pretty intensive render.

Do you really need 5 frames radius? That’s 11 frames to handle for every single frame!
Try 2 and see how much of a difference you can spot. And then, check which algorithm is activated for retiming of you 60 fps in the timeline. It may add up to the load if any version of optical flow.

...I think it's just too heavy for the Mac.

If you are only getting 1-2 fps on 4k material, that is not a problem with the M1 Max. I have the same machine and with the TSNR settings all on maximum, I get about 6.5 frames per second. With lower settings I get about 9.5 frames per second. That is using Resolve Studio 18.1.4 on Ventura 13.3.1(a)

On my M1 Ultra Mac Studio running Resolve Studio 18.5 beta 2 on Ventura 13.4, I get about 32 frames per second with lower TSNR settings. With the highest TSNR settings I get about 11 frames per second.

If you are getting 1-2 fps it's not the hardware. The best solution is create a new project add a single 4k 24 fps clip, then add the same TSNR settings and examine the performance. If that works replace that with a 60 fps clip and try the same thing. At first do not add any other Fx, it must initially be studied using a "bottom up" approach.

Throughout the process, I suggest turning off all Resolve caching, proxies, etc. Otherwise that introduces an unpredictable element. Using this procedure you can first verify the machine and software is working OK. Then you can go back to the main project, copy a small piece of your timeline to a new timeline, turn off TSNR and check the export performance. If that works OK, turn on TSNR and turn off other nodes and try it. Narrow down where the problem is.

This is what I get with NeatVideo in UHD when set to 2 frames radius:

Bildschirmfoto 2023-06-07 um 07.42.36.png (45.56 KiB) Viewed 1029 times

oook, so I suppose, I must learn a lot about importing and caching the files!
As it seems, just dropping the files into the program is not a good idea
So I will check this all out next week and come back with hopefully better results.
Thanks a lot for the input!

#Edit:
by the way, does NeatVideo brings better performance of noise reduction as the build in plugin from DR?

They are pretty close at similar settings.

If I set it to best settings (5 frames, better) and give it about 30% (+ 30% chroma). I get 1-2 frames in quarter resolution

oook, so I suppose, I must learn a lot about importing and caching the files!
As it seems, just dropping the files into the program is not a good idea
So I will check this all out next week and come back with hopefully better results.!

Those values for Luma and Chroma Temporal Noise Reduction (TNR) are enormously high and will result in significant softening of the image. And you should not need such a large radius as Uli says. There is a section on NR in the BMD Resolve Colorist Guide which is well worth doing before you go any further.As well as explaining how to go about it, the tutorial also shows how you can check if you are affecting the image data by applying too much NR.

I have a similar machine to you (14" version) and I can get decent results just using TNR if the noise is not too strong. WIth 4K 24fps raw video in a 4K timeline, I can play without caching to check how much NR I need to apply. It is important to play the clip at normal frame rate to properly check for noise (and flicker). Most importantly it is vital to check on a decent large monitor as it is virtually impossible to discern anything in relation to noise on the MBP screen.

If I can get away with just using TNR then I find it is much faster. Once I add SNR, caching becomes essential so I only use SNR if TNR doesn't do the job. If it is necessary to create cache to get a clip to play, I will do that and then turn the node off until I am ready to export. I use separate nodes for TNR and SNR so I can assess the effects separately and always at the beginning of the node graph for the reason Joe mentions - you can make changes downstream without having to re-cache. If all the clips are shot in the same format, in the same light, with the same ISO and exposure settings, then I can just do NR on one clip and copy the grade to the other clips. Caching in Resolve is quite complex at first but it is worth figuring it out.