Topaz Video AI Beta 3.2.0.0.b

[gregory.maddra]

Hello everyone!

We have another beta release for y’all’s testing.

Downloads
Windows
Mac (DMG) | Mac (PKG)

Changelog:

• Performance improvements
• New updated Chronos Model
• Fixes issues with processing on Intel A770 use latest driver
• Fixes processing failures on Intel Mac (machine crashes still waiting on Apple)
• Redesigned login dialog
• GUI fixes for software updates
• Parameter values are retained when switching scales
• Support for smaller patch updates on windows
• Misc updates

Known Issues

• Stabilization may fail sometimes
• Users with upgrade subscriptions close to expiration may experience login-related issues

Thanks for testing

Please upload problem videos and logs here: Submit files

[david.123]

@gregory.maddra

New updated Chronos Model

Hi, does it have scene detection?

[david.123]

Topaz Video AI Beta  v3.2.0.0.b
System Information
OS: Windows v11.2009
CPU: AMD Ryzen 9 7950X 16-Core Processor              63.139 GB
GPU: NVIDIA GeForce RTX 4090  23.59 GB
GPU: AMD Radeon(TM) Graphics  0.47446 GB
Processing Settings: device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 38.61 fps 	2X: 16.74 fps 	4X: 4.61 fps 	
Proteus		1X: 32.84 fps 	2X: 14.56 fps 	4X: 4.44 fps 	
Gaia		1X: 15.18 fps 	2X: 9.91 fps 	4X: 4.88 fps 	
4X Slowmo		Apollo: 28.6 fps 	Chronos: 15.46 fps 	Chronos v2: 31.88 fps 	Chronos Fast: 36.03 fps 	

[david.123]

I only have Chronos and Apollo available on the interpolation dropdown menu, Chronos fast is missing and there def isn’t a Chronos v2.

[david.123]

In addition to the missing Chronos models, there’s a major performance issue with this build.

Using a 1440x1080 input, upscaling to 1920x1080 with letterboxing, and FPS conversion using Chronos from 23.976 to 59.94, plus Proteus Auto, I get about 3.3 fps. In the 3.2.0.4 alpha, I get about 11.3 fps.

My system info is in the benchmark I posted above.

[suraj]

The chronos model with the beta is different than the alpha. The alpha and all other versions use chronos v2.

[david.123]

Is this the source of the almost 4x performance hit?

Edit: looks like it is, I ran the same processing job as mentioned above with interpolation off and the alpha and beta were about the same speed. So the new Chronos model in this build is causing about a 3.5x performance drop.

So other than incurring a massive performance hit, what does the new Chronos model do?

[ssbroly]

640x480 > 1280x960, 25fps > 50fps.
chronos + proteus auto = 59 seconds (preview 2 seconds)

640x480 > 1280x960, 25fps > 50fps.
appolo + proteus auto = 1m14 (2 seconds preview)

appolo is slower for me so, and from what I see it retains more detail. on the other hand for the previewed ones, around 47% of the first images are frozen.

[20rushtonj (Alaska)]

I would like to comment that the flickering issue with Apollo on macOS with noisy videos is still present in this version.

I have not re-uploaded the files. So please re-use the files I already provided to the dropbox.

@20rushtonj

I’ve noticed that Apollo produces a different result when comparing macOS to Windows.

Details

Description of issue: Apollo Frame Rate Interpolation produces flickering on noisy videos on MacOS. I can not reproduce this issue on Windows. This issue is not limited TVAI 3.2.0.4a. It is also present in 3.1.9

• GPU used on MacOS - M1 Pro 14 Core GPU
• GPU used on Windows - RTX 4090
• Input video details: 1920x1080 at 30fps
• AI details: Apollo (Apollo 8), 1x speed, increasing frame rate to 60fps, Replace Duplicate Frames enabled, Replace Duplicate Frames Sensitivity of 5

I have provided some videos to the dropbox link for the development team to look at.

[20rushtonj (Alaska)]

@gregory.maddra

New updated Chronos Model

I know this can be a bit hard for some AI tasks, but I would like to see a description on what has been changed/improved when a model gets updated.

For example, this is what I would want from release notes:

• Updated X model
  • Performance is expected to be improved/decreased/the same compared to the previous version.
  • In our testing we noticed a significant improvement in scenes with Y attribute
• We have removed Chronos fast from the GUI because of reasons A and B

[rjo]

Topaz Video AI Alpha  v3.2.0.4.a
System Information
OS: Windows v10.2009
CPU: Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz  63.89 GB
GPU: NVIDIA GeForce RTX 4090  23.59 GB
Processing Settings: device: 0 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 26.69 fps 	2X: 10.26 fps 	4X: 2.98 fps 	
Proteus		1X: 20.28 fps 	2X: 9.54 fps 	4X: 2.87 fps 	
Gaia		1X: 12.18 fps 	2X: 8.1 fps 	4X: 3.59 fps 	
4X Slowmo		Apollo: 17.28 fps 	Chronos: 25.74 fps 	Chronos Fast: 26.1 fps 	

Topaz Video AI  v3.1.10
System Information
OS: Windows v10.2009
CPU: Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz  63.89 GB
GPU: NVIDIA GeForce RTX 4090  23.59 GB
Processing Settings: device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 19.45 fps 	2X: 7.14 fps 	4X: 1.92 fps 	
Proteus		1X: 14.85 fps 	2X: 6.13 fps 	4X: 1.88 fps 	
Gaia		1X: 13.02 fps 	2X: 7.84 fps 	4X: 2.97 fps 	
4X Slowmo		Apollo: 17.23 fps 	Chronos: 24.41 fps 	Chronos Fast: 25.96 fps 	

Topaz Video AI Beta  v3.2.0.0.b
System Information
OS: Windows v10.2009
CPU: Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz  63.89 GB
GPU: NVIDIA GeForce RTX 4090  23.59 GB
Processing Settings: device: 0 vram: 0.95 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 25.96 fps 	2X: 10.81 fps 	4X: 2.83 fps 	
Proteus		1X: 20.22 fps 	2X: 9.48 fps 	4X: 2.91 fps 	
Gaia		1X: 12.28 fps 	2X: 8.33 fps 	4X: 3.7 fps 	
4X Slowmo		Apollo: 22.96 fps 	Chronos: 8.61 fps 	Chronos v2: 27.09 fps 	Chronos Fast: 26.3 fps 	

[david.123]

@20rushtonj

I would like to see a description on what has been changed/improved when a model gets updated.

Agreed. How are we supposed to ‘test’ if we don’t even know what the changes are.

Either way, this new Chronos is so slow there’s no point in playing with it further, especially when they won’t tell us how it differs from the other Chronos models.

[malta.tx]

Hi Greg.
I installed it, updated the license and did not download the models.
I attach a screen photo and registration.
Greetings
Processing: logsForSupport.tar.gz…Processing: 2023-04-01-22-35-11-Main.tzlog…

Uploading: logsForSupport.tar.gz…
Processing: 2023-04-01-22-35-11-Main.tzlog…
Uploading: image.png…

[malta.tx]

[20rushtonj (Alaska)]

Also, consistent naming schemes would be nice for the future.

For example, in the benchmark tool, the new Chronos is called Chronos while the old Chronos is called Chronos v2. This implies that the old Chronos is actually the new one when in reality it isn’t.

For me personally, the new Chronos should be called Chronos v3 so we know it’s newer than the old one (and it is chr-3 that in the command line, but not in the GUI)

[20rushtonj (Alaska)]

The OLD Chronos is called Chronos v2 and it uses the code name chr-2 in the command line.
The NEW Chronos is called Chronos v3 and it uses the code name chr-3 in the command line.

The NEW Chronos is available in the GUI under the name Chronos.
You can not select to use the OLD Chronos from the GUI.

The OLD Chronos is in the benchmark, it is called Chronos v2
The NEW Chronos is in the benchmark, it is called Chronos.

[jam7000]

In what Scenario should the new chronos (should be called v3 or v3a if experimental) be used over apollo?

[david.123]

Chronos Fast also not available in the GUI. Detailed release notes could have saved a lot of confusion for us.

[toby.riding]

Here are my Mac Studio Ultra benchmarks. First set, as per default. Second set, selecting the GPU (2nd option in the pulldown menu) and 70% VRAM Settings…

Topaz Video AI Beta  v3.2.0.0.b
System Information
OS: Mac v13.03
CPU: Apple M1 Ultra  64 GB
GPU: Apple M1 Ultra  48 GB
Processing Settings: device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 11.22 fps 	2X: 7.07 fps 	4X: 2.62 fps 	
Proteus		1X: 11.49 fps 	2X: 6.68 fps 	4X: 2.01 fps 	
Gaia		1X: 4.07 fps 	2X: 3.02 fps 	4X: 2.34 fps 	
4X Slowmo		Apollo: 7.14 fps 	Chronos: 9.92 fps 	Chronos v2: 4.85 fps 	Chronos Fast: 6.1 fps 	

2nd Results…

Topaz Video AI Beta  v3.2.0.0.b
System Information
OS: Mac v13.03
CPU: Apple M1 Ultra  64 GB
GPU: Apple M1 Ultra  48 GB
Processing Settings: device: 0 vram: 0.7 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 12.93 fps 	2X: 8.94 fps 	4X: 3.36 fps 	
Proteus		1X: 12.69 fps 	2X: 8.09 fps 	4X: 2.74 fps 	
Gaia		1X: 4.24 fps 	2X: 3.03 fps 	4X: 2.35 fps 	
4X Slowmo		Apollo: 12.15 fps 	Chronos: 8.38 fps 	Chronos v2: 4.47 fps 	Chronos Fast: 6.54 fps 	

[mikmod1]

I noticed that TensorRT Chronos v3 models aren’t loading correctly on RTX 3090 (at least on Win10 22H2 with 531.41 Studio driver), which causes this beta to fall back to .ox model (on the screenshot, files are sorted from newest to oldest, the model files were deleted during the install):

After trying to preview the second time, this beta downloads all the .trt models again, before falling back to .ox.

Proteus TensorRT model works correctly.