Topaz Video AI 5.1 - 5.1.2

I hate to not be able to offer any comments past…version 2.6.4 is what I reverted back to. V3 had issues, 4 had many more, and now version 5…Export As does not work, the GPU is not used past 30%, and any type of preview is dirt slow…you guys need to truly fix your QA. Just because it compiles does not mean it works.


great… you made it difficult to use again and removed looping, and now we have a really stupid and annoying “processing” or “unprocessed” solid background text bubble blocking our view when trying to view footage we are processing or not. There shouldnt be anything on the video especially not things that obstruct the view.
The sidebar covers the work area and you cannot adjust or start the render when the sidebar is open.
I really am not sure how to express how dissatisfied i am with the current version.

Unfortunately it’s not AMD vs Intel, it’s the single-threaded nature of some FFMPEG filters, which is unavoidable.
TOPAZ team cannot invent it’s own video processing engine and de-facto is limited to FFMPEG as the only means of video processing on the market.

So if you want fast video processing, you should get the fastest single-threaded CPU with fast RAM and best Nvidia card you can acquire.
My home setup is Ryzen 5800x + RTX 4070, my work setup is Core i7-10700 and RTX 3070 8gb, and man 3070 is the bottleneck in my pipeline. In terms of CPU, both are equal in most scenarios, but AMD eats less power.

1 Like

In my experience, there’s no FPS conversion without ehnancing.
Usually I go with Aion + Proteus with Manual settings which I can tune after watching Previews.
Yes, Aion takes much more time than Chronos but yields almost perfect results.

This is price to pay for quality, you have other AI viedeo enhancers that are 10x faster but at a moment or another, they do crap.

I would say if the GPU is using 100 percent and the CPU only 3 percent the coding is good. I have seen videosoftware that does just that cuda cuda cuda

If the CPU is using 70 percent and the GPU is 90 percent something is wrong then the CPU is the bottleneck
FFmpeg is not strictly single-threaded. It has the capability to use multiple threads, but the actual usage depends on several factors1:

However, it’s important to note that not all parts of the processing in FFmpeg are multi-threaded. For instance, FFmpeg always has one main thread which does most of the processing2. Also, some codecs or operations might not support multi-threading3.

So, while FFmpeg has multi-threading capabilities, the actual utilization of multiple threads can vary based on the factors mentioned above. It’s recommended to check the specific codec documentation and FFmpeg version for more precise control over threading.


Yes, FFmpeg 7.0 does support multithreading. The 7.0 version of FFmpeg, which was released on April 5, 2024, introduced a multi-threaded FFmpeg command-line interface (CLI) tool12. This is a significant code refactoring and has been a huge undertaking for the project1. It’s particularly beneficial given the increasing core count of modern x86_64 and AArch64 processors1. This means that demuxing, decoding, filtering, encoding, and muxing in the FFmpeg CLI now all run in parallel1. So, you can use the -threads option followed by the number of threads you want to use in your FFmpeg command3. For example, the following command will transcode a video file using 4 threads:

ffmpeg -i input.mp4 -threads 4 output.mp4

This command tells FFmpeg to use 4 threads to transcode the input.mp4 file to output.mp43.

Learn more more



What are some other features in FFmpeg 7.0?


FFmpeg 7.0, also known as “Dijkstra”, comes with a host of new features and improvements1:

  1. Parallel Execution: One of the major updates is the parallel execution of operations for unpacking/packing media containers, decoding, encoding, and applying filters using the FFmpeg command line utility1. This allows for the simultaneous execution of multiple tasks, resulting in a significant improvement in performance and efficiency, particularly when working with large-scale multimedia projects1.
  2. MPEG-5 EVC Support: FFmpeg 7.0 introduces support for encoding and decoding video in MPEG-5 format (EVC – Essential Video Coding) using the external library libxevd1. This new format provides high efficiency transmission and storage of all screen resolutions, supports video with an extended dynamic range (HDR), and panoramic video in 360-degree mode1.
  3. QOA and IAMF Support: The new version has added unpackers and packers of media containers (demuxer/muxer) for the QOA (Quite OK Audio) format and the IAMF (Immersive Audio Model and Format) format1. QOA is used for transmitting audio with loss of quality, while IAMF defines a container for distributing surround sound1.
  4. VVC Decoder: An experimental decoder for the VVC (Versatile Video Coding) format, also known as the H.266 video compression standard, has been added1. This new format provides high efficiency transmission and storage of all screen resolutions, supports video with an extended dynamic range (HDR), and panoramic video in 360-degree mode1.
  5. Hardware Acceleration: The ffplay utility has added support for hardware acceleration of decoding using the Vulkan and libplacebo APIs1. The ability to use API D3D12VA (Direct 3D 12 Video Encoding) for hardware acceleration of decoding of H264, HEVC, VP9, ​​AV1, MPEG-2 and VC1 formats has been implemented1.
  6. Content URI Protocol Support: The new version also includes support for the Content URI protocol used in the Android platform1.

These are just a few of the many new features and improvements in FFmpeg 7.0. For a complete list, you can refer to the official FFmpeg documentation or the release notes1.

1 Like

This is what I told to devs. since days, the CPu and RAM should got that higher and the frame/tile/GOP should be fully given to GPU.

That thing exactly happened with mining time during 2010’. Starting with CPU, migrating to GPU (with diffiulties), mastering GPU, then going to ASIC or FPGA.

ASIC are incoming, Ampere does a great job on this
GPU are allready out with Nvidia ones

Devs have to do with mid. range users = us, a big effort in mastering GPU and workload dispatching

Interesting, ill take a try with the faster 1 threader i have