Please describe what hardware you use Topaz Video AI with and what performance it produces?
My datas:
Victus by HP 15L Gaming Desktop
Processor AMD Ryzen 7 5700G with Radeon Graphics, 3800 Mhz, 8 Core(s),
Installed Physical Memory (RAM) 32.0 GB
Video resolution in->out - Unk
Runs well and process is completed in seconds. Tweaks are sometimes required because autopilot is a bit aggressive but right now no complaints on 2.3.2 or 2.4.0
Your machine is precisely double the speed of mine. I manage about 12 - 13 fps going from PAL SD to Full HD with Proteus on Auto (I did muck around with manual settings but Auto produces the best result). My machine is very modest by today’s standards though:
Ryzen 2700X
16 Gigs RAM
ASUS B450 mainboard
Nvidia RTX2600 Super with 8 Gigs onboard RAM
Intel i5 10600K
Nvidia 3060ti
32GB Ram
SSD’s
I get 35-40 fps upconverting/deinterlacing DVD content to 1080p @25fps.
Only a change of GPU affects the speed.
I have tried increasing CPU frequency, even changing it to i7 11th gen, no change in speed to speak of. Also 16GB RAM is exactly the same. SSD or not, no change.
My cuda cores max out, and that is how I get my speed.
An AMD 5700 gave me about 60% speed of the Nvidia. But we are not comparing like for like.
I would like to know if it would be worth investing in a more expensive, more powerful video card? Will there be such an increase in performance (fps) that it is worth paying the price of a more expensive VGA? I would switch from RTX 3080 to 4080 or 4070ti. In my country these video cards are very expensive.
Yes, that’s why I started this topic. Your fps really surprises me. It seems very little. It might be because of the AI model. Maybe you could do a test with these settings for me?:
Wow! Thank you very much!
Here’s my setup. Not going to say what FPS I’m getting because I didn’t look at it for this test.
Topaz Video AI v4.2.1
System Information
OS: Windows v11.23
CPU: AMD Ryzen 9 7900X 12-Core Processor 31.688 GB
GPU: NVIDIA GeForce RTX 4070 Ti 11.715 GB
RAM 32GB at 5600Mhz with timings at 46-45-45.
I have a 25 minute episode ripped from a DVD. I cut the introduction and credits, then run Nyx Fast with relative to auto settings. Then I run Proteus Relative to Auto upscaling to FHD. Then I convert the images to H.265 with the software encoder and join all the sound subtitles and chapters back in.
Here’s the first run:
Cut time: 0:11:27.71
Starting denoise pass.
time: 0:13:01.83
Running total: 0:24:29.54
Starting enhancement pass.
time: 0:16:51.34
Running total: 0:41:20.89
Starting final pass.
time: 0:16:34.54
Total time: 0:57:59.63
Then I changed out the RAM for some new fancy sticks:
32GB at 6000Mhz with timings at 36-38-38.
Here’s the same script ran again:
Cut time: 0:11:04.27
Starting denoise pass.
time: 0:12:24.31
Running total: 0:23:28.59
Starting enhancement pass.
time: 0:16:08.20
Running total: 0:39:36.80
Starting final pass.
time: 0:16:20.43
Total time: 0:56:01.81
Only the denoise and enhancement passes use TVAI, but those get sped up by a little bit too.
So uhh… better RAM saves me 2 minutes on a ~23 minute show. That will add up the more shows I process.
I’m disappointed. 
For my configuration I bought a Gigabyte GAMING OC RTX 4090 24GB VGA, but with the same settings, I only achieved a maximum of 28 fps instead of 36 fps in fullHD. I don’t know what the problem might be. If I run 2 processes at the same time, then the maximum is 16-16 fps (=32 fps), which is a little better, but far from what I expected. VGA is perfect by the way, it provides brutal performance in games and other software, it does what is expected, so VGA is not the problem. Would the processor difference matter that much (13900k vs. 12900k)?
Good morning.
I am first time using this wonderful program, I am asking how to run the benchmark and where to prevel the log file.
Thanks
Okay. Thank you very much
I don’t know if it has been discussed or not.
Auto setting is SLOW. I have found out that with most movies (that’s what I do) the entire footage is more or less the same throughout. So I find an approx. MANUAL setting, and check it at bright, dark and mid-level parts of the movie (takes a few trials) , then process the movie with that one manual setting! it is much faster this way.
Even with TV series, each season can be done with more or less one manual setting. But setting should be checked season to season.
Did you uninstall and reinstall TVAI after you replaced the gpu?
Yes, now I use manual setting too, because in the meantime I also realized that automatic is terribly slow. Unfortunately, only max. I was able to achieve 36-38 fps with 12900kf+RTX 4090, which is very low for SD to fullHD video. With automatic settings, I reached 24-26 fps with 3080, compared to this, 4090 36 fps is a terribly poor performance, at a multiple of the price of VGA (GPU). It seems that it is not worth making a big jump in VGA (GPU) if the CPU remains the same. By the way, I also tried disabling the use of system RAM in the Nvidia control panel, so only the GPU VRAM could be used, but the performance did not change either. The next step would be to replace the CPU with a 14900k, but I don’t know if it would be worth it…
Yes, that’s exactly what I see, the processing starts at 60-64fps (12900kf) and then permanently drops to around 36 fps. I’m planning to buy a 14900k, but I’m a little worried about whether it’s worth spending the money, time, and energy to install it. I use liquid metal for the cooling, which has to be filed, polished, etc., replacing a cpu is not as easy as in the case of paste. Faster processing is not expected to bring more income either, unfortunately I don’t have that much work.
Unfortunately not. I didn’t think of that, thanks for the idea!
I forgot to write an important thing to you! If you run a single process and it runs at 36 fps, but you want to process more videos, then set the max. process to 2! Thus, you will have 23-23 fps per process with the same settings, so you will encode at 46 fps overall, which is much faster than if you were to encode the two videos separately at 36 fps. You have to pay attention to two things: 1. If you change the max. process, when the videos are already in que, then the second process will not start automatically (this should be developed in the program, the developers should be informed about this), therefore only start the process of the first video, then set it to max process 2 set and only after that put the other videos in the que! 2. With the Max. process 3 setting, I did not experience any further acceleration in encoding (12900kf+RTX 4090), so the 2 threads worked for me well.
I do not understand this:
If I enable two processes at the same time, the encoding works at a total of 45 fps. Why does it run at 36 fps when I only allow one process? Idea?
Because it’as a lame code, not enough trained model, beta seeing aplhas release since at least a month.
End of story 
I wait like a kitten the “professional” release. If they got the troubles we do have, for sure it will turn into a big fail.
Pros don’t want crashes, nor random performances
The 1-year support (update) period has just expired for me, I will not buy it for the next 1 year for now, because the direction of the developments is not going in the direction that I would need. Frame stabilization and interpolation, upscale to 16k. For what??? Meanwhile, the processing speed with a top PC is very-very weak. If this were to be developed, it would be good for everyone.
