Mapping every pixel to spatial and temporal matrices simultaneously requires a lot of thinking. I tried it last night, and didn't sleep a wink.

Exactly! You were using all of your compute and couldn't sleep.

Vegas on the other hand is sleeping like a baby instead of using it's compute.

As I specified in the original post, I have 2 GPUs enabled, The built-in Intel UHD 630 and an AMD Rx580. Task manager shows the complete usage of the device on the left hand side.

So you're wrong,

As I have told you twice, render using denoise, and go through the various gpu engines to check of the GPU activity I have shown.

When using denoise for an encode, task manager shows 1-2% CPU usage , 10-15% Memory usage, 1-3% for the RX580 (main display GPU)

I dont' have an AMD, but considering we get similar performance and your encoding is not dramatically slower I would say it is using a different engine, you're just too lazy to find out what it is. I would not disagree with you that it is slow and inefficient compared to other denoisers, but then you go on to say it's because it's not using the GPU, which it is. It's not using much of the 3d engine that much is true.

Ok, you are right about the GPU but dead wrong about me being lazy, don't know why you have to do that!

I'm running Win10 Pro 2004, with all updates. I used GPU-Z to check and yes the RX580 is being used. Seems like A BUG in task manager prevents it from displaying the correct usage on the AMD GPU.

In addition, my dropdown menu for what to display does not even include Cuda as a choice. I did a search and it appears this is a common problem, I have no idea what the setup has to be to have windows populate the dropdown with Cuda but it is not there for the RX580 or the Intel GPU on my machine.

I changed preferences in Vegas to use the Intel GPU and task manager does show the usage for the UHD 630. BTW: the 14sec clip takes almost 20 minutes to encode using the UHD 630.

As I specified in the original post, I have 2 GPUs enabled, The built-in Intel UHD 630 and an AMD Rx580. Task manager shows the complete usage of the device on the left hand side.

So you're wrong,

As I have told you twice, render using denoise, and go through the various gpu engines to check of the GPU activity I have shown.

When using denoise for an encode, task manager shows 1-2% CPU usage , 10-15% Memory usage, 1-3% for the RX580 (main display GPU)

I dont' have an AMD, but considering we get similar performance and your encoding is not dramatically slower I would say it is using a different engine, you're just too lazy to find out what it is. I would not disagree with you that it is slow and inefficient compared to other denoisers, but then you go on to say it's because it's not using the GPU, which it is. It's not using much of the 3d engine that much is true.

Ok, you are right about the GPU but dead wrong about me being lazy, don't know why you have to do that!

I'm running Win10 Pro 2004, with all updates. I used GPU-Z to check and yes the RX580 is being used. Seems like A BUG in task manager prevents it from displaying the correct usage on the AMD GPU.

In addition, my dropdown menu for what to display does not even include Cuda as a choice. I did a search and it appears this is a common problem, I have no idea what the setup has to be to have windows populate the dropdown with Cuda but it is not there for the RX580 or the Intel GPU on my machine.

I changed preferences in Vegas to use the Intel GPU and task manager does show the usage for the UHD 630. BTW: the 14sec clip takes almost 20 minutes to encode using the UHD 630.

 

Found this old thread as I am grappling with the same issue - slowness of Vegas noise reduction. One thing about CUDA - it's Nvidia only technology - so you would not see that on Intel or AMD. They would use something else like Intel OpenVino or OpenCL on AMD to accelerate compute.
This still does not explain how come Vegas noise reduction plugin is so slow - and to add insult to injury, I find Neat video (trialing) also producing better results. Same slowness with the (suposedly AI) Style transfer plugin. This seems like a first attempt on Magix's side to dabble into the area with no in house expertise.

Neat Video is a very mature piece of software and I think had its roots in even older noise reduction software for photos, so I'm not surprised it has better results and better performance. Hopefully Magix keeps developing the tool and v2 runs faster. I didn't want to wait myself so I bought Neat Video last winter.

This still does not explain how come Vegas noise reduction plugin is so slow 

In order to back up such a claim, you would have to provide fair test comparisons to other denoisers, both standalone and in other applications. I issued the challenge above, but of course no takers.

Play around with NLMeans, QTGMC, Hqdn3d, and other command-line solutions

Yes, I would expect Neat to be a little faster, but just how much?

Yes, I would expect Neat to be a little faster, but just how much?

Factor of 6 on my system.

I had tested denoise before. I had also used the deflicker feature in Vegas which combined with motion tracking was able to do a sort of temporal noise reduction. However it was unstable and often crashed on render so I stopped using it. I had also tried the freeware denoisers as Voukoder has them available upon output (no command line necessary.) Testing is blind, though (no preview) so it's hard to make a fair comparison as trial and error is involved.

The Vegas denoise Fx was so slow that repeated tests are impractical- a short <1 min clip will lock up the machine rendering for 30 min...

Even for playback I can't see Vegas denoise in motion way under <1fps. Neat Video at best/full is >3fps. That makes it hard to compare quality before rendering. The NeatVideo looks much better in the preview window, getting rid of most chroma noise while preserving details. Denoise has blotches remaining.

Denoise Fx uses about 30% CPU and all of my GPU's CUDA capability:

NeatVideo uses more CPU and less CUDA:


Onto the 30 sec. clip render test
(Source: noisy Canon T3i HD 1080p 24p file rendered MagixAVC 1080p 24p)

NeatVideo (temporal & spatial denoiser): 4:28; avg. speed 3 fps
Denoise Fx: 26 min (in reality 30 min+; avg. speed .46fps

Neat Video CPU render:


Denoise CPU render:

In the resulting files, Vegas did an okay job, but Neat Video's result was cleaner with greater detail and looks far better in motion thanks to temporal NR.

Original:

Vegas:

NeatVideo 5:

If you want quality denoising that is fairly quick (at least compared to the above numbers), you might want to try HappyOtterScripts.

Here's a demo using the same sample footage with one of the two HOS tools that support denoising.

Here is a link to both the original and rendered file as well as the above demo. (To view, I'd suggest downloading. Don't use the DB player--it's terrible). https://www.dropbox.com/sh/u5hp37sfcjhvbre/AADKMrBBldHG328iPtXPxTuza?dl=0

The file was directly rendered from AviDub using x264 at the default 23 CRF setting. The temporal denoiser MDegrain3 was used which looks at 3 frames before and after each frame. There is also an MDegrain2 option which looks at 2 frames before and after which produced somewhat quicker renders of 18 fps.

When rendering to the lossless codec MagicYUV, the following speeds were obtained. For MDegrain3.

and MDegrain2 which was getting pretty close to real-time.

You can take a look at my fairly modest system specs in my signature.

@wwaag Thanks for running those tests- those are very impressive times for noise reduction! I downloaded the result and it also looks good on part with NeatVideo. Gamma looks a bit different, though.

I went into Voukoder and did a hqdn3d test (4333) and it rendered in about 2:30 but noise reduction wasn't enough. I did a second pass with stronger settings (6656) It takes about the same time. Almost no GPU activity (including CUDA) and CPU is maxed out but it is faster. Wwaag's results look much better but both are far better than the source file.

 

@wwaag & @RogerS Huh!! As we’re Ole Timmers ‘round these parts: how does it stack up against John Meyers, one might say the obi wan kenobi of Noise Redux, approach to noise reduction?

@Grazie

The denoiser used in HOS "IS" the John Meyers approach. Specifically, the Avisynth scripts used in AviDub for MDegrain2 and MDegrain3 processing are directly from him. That's the reason they compare favorably and are arguably of better quality than what NeatVideo delivers. Although a customer for many years, I must admit that I've never been a fan of NV and its photo predecessor Neat Image because of their tendency to produce a "plastic" look in the final output due to an emphasis on spatial processing. As I recall, that was one of JM's observations as well.