Topaz Video AI v3.0.10

In my 30+ years computer experience, it’s almost always the PSU. Crashes, black screens of death, all signs of a weak PSU. People simply need a good PSU. Wattage is only one thing; stability is equally, if not more important.

And RAM. MicroSoft said this 20 years ago already: ‘All problems are solved by adding more RAM.’ While in jest, of course, also true in many ways: the less disk I/O has to be performed (due to excessive swapping), the lesser your chances of a crash.

I have 64G myself, and a high-quality PSU. TVAI does many things wrong, but it has literally never crashed on me.

1 Like

I generally do 2 or 3 upscales at the same time. I have my processes set to 1 since I just open another instance of TVAI. When I do do it’s generally a very small total time difference to a single encode. When I do 3 that takes about the same time as doing 2 back to back. I rarely do anything but 480 to 1080. If I was doing 1080 to 4k that be pushing my RX 6600 or 3060 Ti since they are only 8GB cards but my 3060(12gb) could probably handle it.

Here are 3 instances running of the 3060 Ti. The processor is a 3900x.


I have the Cooler Master MWE Gold 850 - V2. My harddrives are ssd’s. The smallest one has 130 gigabytes left.

Indeed. Memory usage can also be a little fishy as the graph may not max out, instead a swap file on C: take over, and everythings looks well but the system slowing down quite a bit.

There is this setting in Windows 10 and 11 that makes changes in the GPU usage. It also changes how Task Manager reports the GPU usage. It’s seems much more realistic when compared to other software that reports GPU stats.

While changing the setting to On did not seem to make any difference in execution speed of TVAI it does show the GPU is being used which is what I’ve always seen by the GPU power draw.


This problem is still not fixed:

yes the new radeon 7000 series has av1 support, also would be interesting if they added support for smart access video, I barely see anyone talking about this feature with the new radeon 7000 series combined with amds new zen4 cpus, its basically amds equivalent to intels quicksync. would be cool to test these two new technologies to help benchmark this new card as well as hopefully speed up export times. Also is their any plans to incorporate either nvidias tensor cores or amds new ai accelerators to also boost export times??

So, I tested 3 concurrent 4k upscales, different scenes/trims selected from the same video. Combined was 2.4 fps. In comparison, exporting a single stream was 1.4 fps, so it’s a considerable gain.

GPU processor utilization was 50%, CPU was 56%. So, in theory there is lots of room for additional streams.

My system is i7-7700k, rx6800xt with 16GB, 48GB RAM.

Just want to add, TVAI v3 is generally about 10% slower than v2 in my case.


A suggestion: it will be awesome if you make an instant preview while we set the parameters to Proteus ( or any other model), I mean you can make it apply the setting to just one frame instantly that way we can get an idea about the results like sharpen and reduce noise without click Preview every time.



1 Like

There are 2 other places in a modern system that could cause slowdowns that I don’t know how to monitor. One is the PCIe bus lanes with all the traffic that runs on them. The other is the CPU internal cache.

Still only using 0-30-40% of my 6700XT’s horsepower. Up and down up and down.

CPU 38 % GPU 43 % with the Alpha on MSI RTX 2070 Super atm.

So would you say this is more successful than running three together in 1 window/instance? Never thought of that. :+1:

I couldn’t tell you as this is the method I have been using since VEAI 1.1 . I do this with a lot of programs, the ones that don’t block doing this. For me if I have a GUI crash like we have had in the past, if I lose one GUI the others are still there and running. I still have a visual idea of where my progress is and have GUI control over those other video.

The PSU is large enough, but unless you’re trying to do a very small sized video, I suspect that you may be running out of room. With all the temp files, etc,. you may not have sufficient space.

Go into properties and do a cleanup to see if you can’t get more free space. I don’t know what format or video size you’re using but it can add up.

Good luck…

additional review

I converted MPEG-TS with DDV3 and tested a scene with a lot of motion

DDV3-2.64 3840x2160
The image quality is the best, but there are deinterlaced parts and afterimages

DDV3-3.0.5 3840x2160
Deinterlacing is good, but there is noise overall

DDV3-3.0.10 3840x2160
Same as DDV3-3.0.5

DDV-3.0.0-5 2880x2160 (Early Access)
Good deinterlacing noise removal

MPEG-TS-DDV3 created with 3.0.0-5 gives the best results when using Chronos


Confusing. You list out four versions of the program then say you used the Chronos model on three of them. Your comments on each version suggest you’re using one of the many deinterlacing models. You don’t say what model. (It cannot be Chronos because that does not deinterlace.)
What are we supposed to understand from your post and the example images?


Good Comment. -
The best enhancement results start with getting the cleanest input. Deinterlacing and denoising are the primary problems.

The need for effective deinterlacing is paramount. Knowing how to deinterlace, how to identify the various flavors of interlaced video and select the best methods for processing them is essential.

We need to form a separate discussion dedicated doing this using TVAI 3.x on this topic.

Unfortunately, after saying that, it may take a while to break off enough free time for me to start one. However, if anyone else cares to start one, I’d be glad to participate when I can.

Happy New Year, to everyone! Lets hope that 2023 make the world a better place.


It would be wonderful if TVAI had some analysis routines that could test the individual user’s system and give feedback on the maximum number of multiple processes it recommend for that system.