Performance increase

Hi, how can I increase the video processing performance, right now I am improving with the same resolution and FPS as the original file. using , progressive, artemis - denoise/Sharpen, with strong halo.

Output h264 high Nvidia, MOV, auto bitrate, no audio.

In preference I have Ai processor in car.
As I say, I have not changed resolution or FPS other than 1920x1080 FPS 25.

I mark 5h
My cpu usage is 38% i7-6900 3.20ghz
And GPU between 5% to 24% RTX3060

What can I do to improve processor times…
My video is 1h and 12 min.

in the preferences enable multiple instances and then process the first half and the second half of the video via trim/crop simoultanously!

1 Like

Like this or get a faster CPU.

A CPU bottleneck is likely confirmed.

https://www.reddit.com/r/TopazLabs/comments/youdh7/how_long_to_upscale_1_40_minute_video_rtx_4090/

Quote from user “elementalSG” on reddit.

"I’ve found that Video AI’s performance with certain upscaling algorithms (esp Artemis) more bottlenecked by the CPU than the GPU. For example, my gaming computer with an older i9-9900K but an RTX 3090 somehow upscales slower with Video AI than my Ryzen 9 3950X with RTX 2080 Ti. And oddly my server Ryzen 9 5950X + GTX 1080 Ti is only like 30% slower than the other 2.

Seems like you will want the fastest CPU to go with the 4090 if you want the best performance with Video AI off the data from my two different rigs. At least that’s what I’m experiencing."


Keep in mind that the 9900K has PCI-E 3 while the 3000+ ryzen CPUs have PCI-E 4.

One of the Programmmers did mention that PCI-E bandwidth does matter.

Since TVEA is a GPGPU workload, this is true.

So maybe when you invest in a new system, get a Ryzen 7000 CPU (PCI-E 5) and a RDNA3 GPU (PCI-E 5), this should be the fastest you can get in January 2023.


And here is a user with a 4090 (PCI-E 4) + Ryzen 7950X

I tried this on Mac Mini M1s – and it just made each process longer. Resulting in no advantage but actually a disadvantage. Perhaps it is different on a PC.

Forgive me for posting this again, but these were the results – unless something has changed in the 3.0.0.2 version.

the seconds per frame increase a bit with each added instance but the overall time of processing will be lower.

Ahh - stupid me. I misinterpreted my own results. I used copies of the exact same file for each instance (with the same settings). And I ran this for about 5 minutes.

So if this is true - then there is an advantage, but I have to wait for all files to complete (if using different videos and not sections of the same video).

If the software can automatically do this in the background – then you definitely would have something. Very similar to how Apple’s Compressor would chop up the file, run multiple instances, based on your # of cores and then put it back together again.

Only issue is if you are using SSD or a mechanical HDD.

1 Like