That’s why i finally bit the bullet. It sure isn’t the new “UI”.
You also get those spikes, that’s not a good performance at all, it should be at 100% all time.
And go back to the image of when I was using TVAI 3.2.0 and what percentage was it at and did it have any major spikes ?
1 Like
The 3.x software, cluttered (tiny) UI, mega cluttered options and so on is still just SO much worse than the old ones. Instead of selecting model, parameters and size quickly, everything is hidden away in a chaotic mess of submenus. You’re basically being forced to use all the pointless “assistants” of 2.6.4 with no escape.
The new trim mode is a small but prima example of what is wrong with Topaz’s UI mess. Instead of setting in and out markers with two clicks you use a absolutely unnecessary submenu with no feature benefit at all! Multiple sections would be a justification for it! But no, it’s still just one in and out marker (or i’m stupid). You just need more clicks to do the same very limited thing. And if you wanna change it, you can’t just move 2 markers on the fly, you have to go through the whole clicking orgy once again.
There is still no pause/resume function. You still can’t save a session manually. Even the load last autosave seems to be missing?! Maybe it’s hidden in the dark abyss of the sub menus and i just didn’t find it. You still have too few and can’t edit hotkeys. NOTHING (except models) has improved, everything is just more bloated and cluttered.
And the other all in one softwares aren’t much better.
It really is just the speed. That has improved so much that sadly there is no way to avoid the 3.x workflow nuke any longer. The new models in the old, simpler UI would be a dream. Thank god Topaz seems to be better at those than making UI’s
Generally all the newer all in one softwares like Photo AI are just a workflow wasteland compared with the more barebone older ones. Trying so hard to be convenient and think for me, that they are much more inconvenient to use.
1 Like
Haha maybe stop trying man, they obviously can’t look to the right… 
1 Like
Ha ok. Yeah, that’s right, it’s a bug. The installer should check if there is already an icon to know if it should add one or not. The other topaz softwares (gigapixel, denoise, mask, Video enhance AI (until 2.6.4) and Jpeg to raw) don’t have the problem
According to your taskmanager your CPU is running at 94% utilisation. The program isn’t holding back your GPU.
1 Like
Topaz Video AI v3.2.0
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 9 5950X 16-Core Processor 31.906 GB
GPU: NVIDIA GeForce RTX 4090 23.59 GB
Processing Settings: device: 2 vram: 1 instances: 0
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: ERR fps 2X: ERR fps 4X: ERR fps
Proteus 1X: ERR fps 2X: ERR fps 4X: ERR fps
Gaia 1X: ERR fps 2X: ERR fps 4X: ERR fps
4X Slowmo Apollo: ERR fps Chronos: ERR fps Chronos Fast: ERR fps
That’s because of the encoding, I’m using H265 (Intel).
Also, look at the utilisation and VRAM used:
And of course, the temperature:
Whatever you say, this programme isn’t using shit. And of course, the 4000 series GPUs are performing much less well…
This is all because the developers are reducing the tiling and VRAM usage constantly to make their software more compatible with low gamma GPUs instead of focusing on improving performance on high end GPUs like the 4090.
How can you justify such minuscule tiling, it makes no sense whatsoever, which is why this programme is so excessively slow and unoptimised.
Seems like they only care more about selling their software than they do about providing a good user experience.
PS: Any other AI you see out there will use 100% of the GPU, causing the GPU to get very hot, which is expected behaviour. You mentioned SwinIR earlier, that’s funny, because SwinIR requires a lot of VRAM from the GPU, and you have a 3070 with 8GB of VRAM. I have a 4090 and to upscale from 1080p to 4k I have to set a 512x512 tiling to do so because otherwise the 24GB of VRAM is exceeded.
PS2: How can you justify I’m getting 2fps for an upscaling to 4k and interpolating with Apollo to 60fps?
TVAI also uses the cpu, and at 94% utilisation, you will have a bottleneck occurring.
1 Like
Really? 
Dude, I’m using a i9-13900k…
What I find funny is the first 3 images you quote in this post that I posted initially are in fact another AI and not TVAI.
I’m going to end this conversation here as it is now pointless.
FYI: If you understood anything about tiling and various model platforms you have solved your problem with what you have wrote but just can’t see it thro lack of knowledge.
2 Likes
Yes really. The 4090 can bottleneck even a 13900K. Your own task manager shows it.
VEAI isn’t fast enough to bottleneck the CPU. I use CAIN sometimes, on its TensorRT implementation and I get 200fps at 1080p and I can tell you for sure that I don’t even get bottlenecked with that.
So you don’t have nothing to say about what I said, ok that’s good to know.
And yet your CPU is hitting 94% utilisation according to your own screenshot.
I’m sure the devs will take your advice about everything the’re doing wrong onboard…
What part don’t you understand that it’s because the encoding? 
That’s mainly being used by FFmpeg dude.
And that’s absolutely not a justification for why I’m going 2fps with a 4090. I use other AIs like CAIN, in their TRT implementation, I get around 200fps on an interpolation at 1080p. The process encodes the video in the meantime, and the processing speed of the AI is not affected. How do you explain that AI not causing the same effects as VEAI? Exactly, that AI is well optimised and knows how to use the PC’s resources well, VEAI is incapable.
I gave CAIN as an example because it is the fastest, but I could give other examples such as RIFE, ESRGAN, RealCUGAN etc, even SwinIR, which is slow af, but uses the full resources of the PC.
The worst thing is that this is paid software, which costs a lot of money, and those AIs I mentioned are open source software that anyone can use for free. Sadly, there is no open source AI that matches in upscaling quality for IRL and CGI content, so you have to deal with this slow, unoptimised thing and deal with a community that defends the programme’s dismal performance because they don’t have a clue.
I’ve said all I have to say, it’s up to you to continue living in a fantasy or not.
I’d love to see the devs replying here about the poor performance on a 4090, I really want to know what’s their justification.
1 Like
Topaz Video AI v3.2.0
System Information
OS: Windows v11.2009
CPU: 13th Gen Intel(R) Core(TM) i5-13600K 31.747 GB
GPU: Intel(R) Arc(TM) A750 Graphics 7.9063 GB
GPU: Intel(R) UHD Graphics 770 0.125 GB
Processing Settings: device: -2 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 5.90 fps 2X: 3.72 fps 4X: 1.16 fps
Proteus 1X: 5.38 fps 2X: 3.04 fps 4X: 1.00 fps
Gaia 1X: 4.45 fps 2X: 2.76 fps 4X: 2.33 fps
4X Slowmo Apollo: 6.07 fps Chronos: 2.57 fps Chronos Fast: 4.83 fps
TVAI uses a custom FFmpeg which handles all the AI processing too not just the video encoding. The TVAI app is mostly just a frontend gui. If FFmpeg is what is heavily utilising your CPU, then that is AI processing, not just encoding.
I’m sure they’ll be other optimizations over time, but at this moment your CPU is the bottleneck and why your GPU is under utilised now. That’s all I’m saying. I’ve never said that the current speed on a 4090 is “good” or can’t be improved.
I used to watch videos from an external media player. It was automatically adjusted to different fps values. But, now I watch videos from my computer via HDMI. I’ve noticed that all videos that have a different meaning than the 60Hz TV clock speed bounce and flicker a bit. Best of all, without jumping, the TV shows a video that has fps equal to the frequency of the TV Hz. 60=60! In the future, I plan to buy a 120Hz TV, so I make all the videos 120fps…
