Processing Times - Newer Versions Faster?

I am contemplating purchasing another year of upgrades. Most of what I upscale is 480p by 2x, which typically is about 2 hours of processing for a 50 minute video, which I am happy with. I still use VEAI (2.6.4). I’ve tried Video AI (3.0.12, last update I received before my upgrade period expired), but the processing times were about the same, and I liked the VEAI interface better.
I recently tried to upgrade a 1080p video to 4K (or 2x), about 1 hr 50 minutes length, and both VEAI and Video AI said this would take about a day to process.
Video AI 3.1.0 says that this version has faster processing. Anyone have any idea “how much faster” could be expected? My computer is an i9-11980HK with NVIDIA GeForce RTX 3080 (Laptop). If I could expect, say, 25% reduction in processing time, I’d probably upgrade.
I can’t run benchmarks because the current version(s) I have do not support benchmarking, to my knowledge.

Hard to say exactly with a laptop, but the speeds are much better.
I have a 3080 ti and I just tried a 30 second clip on 2.6.4 and then on 3.2.6 with all the same settings.
It took about 2 minutes on 2.6.4 and 30 seconds on 3.2.6.

If your laptop is anything like that, you’ll get more than double the speed on most models.


They have improved performance, of course. But not enough when you do processes like “Upscale and interpolate to 4k60fps” from a 1080p video. I don’t know why it’s so exaggeratedly slow when other programs do it at higher speeds.

You can test the performance for yourself.

You can just download the latest version of TVAI, leave it in trial mode, import a video, and create a preview (10 seconds, 15 seconds, etc) and compare it to older versions of TVAI or VEAI creating a preview of the same video with the same length.

You can find all the downloads here: Releases - Topaz Community

1 Like

well, based on Topaz since v3.012 there were two iterations that had performance boost:
v3.1 including on RTX 3080 (from 334 sec for 10 sec video upscaling to 4K down to 174 sec).
and on v3.2, another performance boost.
how about you download the latest version (Demo mode) encode a 1min clip from your usually upscaling library and compare the results for yourself compared to your v3.011? that would be the best real life results for you.

Because they do much better job then other programs and that comes with a price, performance.
more time you spend processing, better results you get. it’s very simple.


I don’t think that’s a valid reason to justify it. The possible reason is somewhat more technical and has to do with VRAM tiling. In fact there is an “official” image that indicates this:

Honestly, although I don’t trust the source of the image (I’ve tried to look for now and couldn’t find) nor the arrogant Topaz-hating rabble who shared it with me, it makes sense to think that low VRAM tiling would cause these slowdown problems. What would be needed would be for a developer to explain in detail the reasons why the program is so slow and why they can’t speed up the performance.

1 Like

Well, you are more then welcome to be part of the R&D and delivery team to find more details on why the process is slow and find a creative way making it much quicker that Topaz developers have not figured it out yet…
But the fundamentals remain remains the same, like with any other things in the tech world.

better quality = more cost (more time, more energy, more expensive, etc.).

There is always a trade off of something. Either speed for quality or quality for speed.
You can’t hold the rope at both ends. or eat the cake and keep it whole.

You need to remember, most of us here, those who picked Topaz are here predominantly because of the quality, That is our Primary objective, or we would be elsewhere. there are many tools out there claiming to do the same, some are likely much faster. but none are any close of being as good quality as TVAI, at least not that I have encountered with.
The Speed, is our secondary objective.
If speed was our main motive, we would be elsewhere, with more simple tools or video editing tools, etc. that do it quick but dirty.

Finding the balance (Speed vs. Quality) is the tricky part, the art. and that is what Topaz are doing & improving every PI, on every Iteration.

P.S. I don’t think low VRAM is the cause for slowness. if i had an RTX 2080 GPU with 100GB of VRAM (assuming there was such thing), I don’t think the process would go much faster, for sure not quick as a RTX 4090 with 24GB.
Maybe it would be slightly faster than traditional RTX 2080, but not to the level of “Solving the speed problem”
and if it is due to how VRAM works today, Topaz has little way of influencing, as this is to the hardware vendors or Driver, OS , DirectX , etc. vendors to change. TVAI is/running at the application layer, not the system/core layer.


With the old version I processed a movie in maybe 9 - 12 hours. Now with the new version it is done in 4 - 7 hours. It will depend on your individual hardware. :slight_smile:


I am not a programmer? Are you? At least I tried to find a logical explanation since the developers of this program never said anything about the lousy performance, and I repeat, even if the source of that image is dubious, it makes more sense than any other theory. Since other AIs with TensorRT are much faster, and in theory the VEAI models are TensorRT according to the people who shared the image with me about the tiling. I use another video interpolation AI called CAIN, basically the same as the Apollo model. But the speed difference is abysmal as in their version of TensorRT with hardware encoding I sometimes exceed 200fps in a 1080p resolution video. Here in VEAI I can’t even reach 50fps despite having selected hardware encoding. I also use RealESRGAN, specifically a model designed for anime, and the speed at which it scales from 1080p to 4k is usually around 15fps, Topaz Video AI doesn’t even reach that speed, it stays between 3 and 9fps. Quality doesn’t mean slowness, another interpolation model specially trained for anime like “GMFSS Fortuna” makes the interpolated frames look native and although it is much slower than CAIN (15fps) it is still much faster than VEAI.

Well, these other AIs/models I mentioned use a much higher amount of VRAM. RIFE which is a video interpolation AI based on optical flow, just like Chronos, consumes more VRAM and is a bit faster.

I’m not here to discredit the programme, but considering that most of us use this programme to enhance long running footage (or at least that’s what I do) it doesn’t make sense to have to wait several hours for it to finish (it’s not even possible to pause/resume the process), it should be much faster.

Also, if you look at the CUDA graph in the task manager there is a lot of high/low usage peaks and generally low GPU usage, at least that is the case on the 4090. The other AIs I mentiones just use the 100% during all the process… And this is noticeable both virtually and physically, through the sound emitted by the graphics card.

Perhaps it is simply that the programme has a different internal structure from the rest. The truth is that objectively there is nothing to talk about, as long as the programme works it’s all good.

The thing is that I come from a discord server of people who are enthusiastic about video and image post-processing with artificial intelligence, they have a lot of advanced knowledge and they even made models of ESRGAN or even SwinIR if I remember correctly. And well, these people are very arrogant, proud and egocentric. Their arguments towards Topaz VEAI are “skill issue”, “scam”, “trash” and what I mentioned about tiling. And well, apart from their clowning and immaturity, they seem to have a point since they know a lot about artificial intelligence and it makes sense to think that it’s a bad tiling management in the VRAM. Although after leaving their server, I’d rather just ask, find a logical reason than judge Topaz like they do.

Then why are you still sticking with Topaz if the others you mentioned sound so wonderful?


Because there is nothing better at the moment for upscaling/denoising/stabilising IRL videos.

Exactly my point I was trying to convey earlier…
Better Quality = comes with costs.

You can’t hold the rope at both ends.


Why do you insist on such “copium” reasoning?

That reasoning doesn’t tell me anything, what’s more, don’t you think it should be the other way round? If you pay for something (moreover, so expensive) it should comply with speed and quality, not only with quality, because if it is so slow the programme loses a lot of value.

I gave you plenty of examples, and you just keep saying the same thing over and over again, if you don’t have anything better to say, don’t keep insisting.

listen my dear friend, I think we are going in circles here.
let’s end this debate and agree to disagree.


Thanks to all for the replies and lively discussion! I think that if I could halve the time needed to upscale 480p alone that would justify buying the year upgrade (currently with a discount from Topaz!). Topaz does seem to be committed to consistently improving the software, and I do use it quite a bit. SD video upscales horribly to 4K on a 70" TV, therefore Topaz software is invaluable in order to make that SD video viewable. If I could upscale a 1080p video of about 2 hours length to 4K in under 18 hours, that would be a big bonus. At this point the upgrade seems to be worth it.


@alan.mckown - yeah the upgrades have definitely improved things and I’m glad I made the investment. :slight_smile:

1 Like