File split / recombine to increase VEIA throughput?

Topaz Video AI General
1

I use VEIA daily to upscale original content for streaming at 720P. Artemis MQ with .5 grain works well for 720x480P sources. Using a 24-core AMD 3960x with RTX2080i GPU, upscaling averages 2.5x real time. (1 hr content takes 2.5 hours.)

If I split the hour of content into three 20 minute videos, and run three instances of VEIA, the hour of video is done in 56 minutes, since one instance uses slightly less than a third of available processing power. Running three instances, the PC becomes a handy space heater, an added benefit in cold weather.

Any chance of offering a file Split/Recombine option for machines with extra cores available? (I can do it manually with an editor, but automation would be super useful.)

Or maybe this upgrade opportunity has been addressed, and I missed the post?

Regardless, thanks for fielding requests like this.

1 Like
2

Yeah it’s a little inconvenient.
I often scale content to 4K and it takes awhile.

I don’t split a video beforehand but I do keep the source file on a different drive from where I’m aiming to save my outputs.
I simply run two instances of VEAI and tell each of them to output into an image format and give them their range on how much to do. An example for a 1000 frame video is the first instance runs 1-500 and the second isntance runs 490-1000.
I keep a 10 frame buffer to overwrite just because.

Afterwards I use ffmpeg to export the audio from the source video and stitch that together with all the images back into a video.

Probably only saves me 20% of the time of one instance of VEAI when doing 4K content. But is just a bit of extra work on my own part.

1 Like
3

Thanks for the great idea. I’ll try the same. The upscaled video, containing mastered audio, is being added to the final timeline anyway. A few seconds to edit in an extra segment or two, and it’s ready for output.

Big timesaver! Thanks again.

4

I tried. It will slow down and the estimated time is not accurate anymore. More GPU memory used = slower.

1 Like
5

Hi Viktor. Thanks for your comments. I ran a second test, to see if/how speed has changed since last year. A two-hour video was copied to two different filenames. One instance of VEIA took 330 minutes to finish. Then, upscaling two one-hour segments at the same time took 170 minutes.

So that’s 165 minutes per hour - single instance, vs. 85 minutes per hour - two instances. There were no heat issues, and both instances hovered around .09 spf from beginning to end. This is the same ballpark as the Dec 2020 test.

6

This would be true if your video is not 1080p + not Gaia models. 480p/i is fine with multiple instances. Artemis models work great with multiple instances as long as CPU usage is sufficient. And for best picture quality, I use TIF. And TIF depends the most in SSD or M.2. But 25m video costs me around 800GB of image sequence.
Still depends on how you use VEAI. But split the footage is a great thing to do if your PC is capable of. But the more instances you run, the higher chance that your system will crash or have frame error when you export.

1 Like