Processing speed goes down after certain number of conversions


I am currently trying to convert a series of videos. I notice that after one or two conversions of videos of same resolution, and similar lengths and bitrate (around 15 min) using the same settings each time (for reference, upscale to 4K from 1080, progressive, Theia Detail, Sharpness 40, Grain amount 5, size 1), the processing speed goes down from initially 0.3-0.4fps for the first video I work on to 0.1fps for the next (or two afterwards) and never recovers. Even after restarting the application the processing speed does not return to around 0.3-0.4fps. The only way to get this higher speed is to restart the computer, reopen the application, and apply my settings to the video then I get 0.3-0.4fps. I would prefer not to have to restart my computer so often.

I am using the following:

  • Topaz Video AI version 3.1.6
  • MacBook Pro Apple Silicon M1 (2020)
  • 16GB RAM
  • Mac OS Ventura 13.2

Performance Settings:

  • Use GPU (Apple M1)
  • Max. 100% memory usage
  • Max. 4 processes
  • Low power mode setting off

Is there something that can be done or needs fixing such that I don’t have to keep restarting my computer every time I want to convert another very similar video to one previously converted? The fact that a computer restart rectifies the issue suggests maybe something is possibly not being ‘reset’ or ‘cleaned up’ after each conversion?

Thank you for your attention.

I am going to reply in this post, because I am encountering the same issue, but from a different cause.

As I have V2 of the application still that I use, anytime I open the V2 and then (after closing V2) and opening the latest V3, V3 will process at 1/4 of the speed.

I can do multiple tests and an upscale in V3 with no issue, but the second I open V2 it kills V3 processing time, regardless of restarting apps. It does not cause any slowdowns in V2, it keeps going as normal.

Like above, only restarting the PC fixes it, so anytime I use V2 at all, I have to make sure to restart the PC before trying to use V3 again.

I wonder whether this is also linked…I have noticed that when I receive an ‘Out of Memory’ error from processing a video, trying to process another less intensive video produces ‘Out of Memory’ without even trying to process. Restarting the application does not resolve this. I actually have to restart the computer again! And then I can process the aforementioned less intensive video without problems. Is there some missing clean-up logic?

@ebee Thank you! We are looking into this :slight_smile:

Hello and thanks for reporting this issue,

It appears that in some situations the app is not completely releasing the memory used to process videos, and a full system restart is required to fully release the memory held by Video AI. This is definitely not intended behavior, and I will check with our dev team to see what our timeline is for addressing it.


Our developers are suggesting to reduce the number of parallel processes to 1-2 in the app’s preferences menu. This will reduce the amount of memory required by the app and prevent reduced performance over time.

Hello, thanks for the reply and for looking into this. Unfortunately, the problem still persists even when I have had maximum number of processes set at 1.

Thanks for trying that setting. Is this still on version 3.1.6? I’d like to check if the memory management improvements in 3.2.0+ would help in this case, as it does seem like the app is not properly clearing out of the system memory even after quitting Video AI.

It occurs even on the latest version 3.2.2. Also, I notice on the first one or two runs after a restart certain components such as the Apple Neural Engine appear to be used more indicated by higher temperature and there is much fan activity, but on subsequent runs where the processing speed goes down, the said components appear not be utilised to the same degree as per lower temperature on these components and lack of fan activity. This happens regardless of whatever settings are made to CPU/GPU, parallel processes, low power mode.

memory leak? Memory leak - Wikipedia