GPU processing

The program has been stable for me. I have a 3070 now and was going to purchase a 3090.
When will Video AI use the GPU to its full potential?

Could you elaborate as to how it’s not being used to it’s potential?

I encode 3 different streams simultaneously on my 12GB 3060 and have almost full RAM and GPU utilisation (as well as CPU utilisation).

Have you thought about running streams in parallel or if a single file, splitting it up?

What fps are you getting?

Having the same issue…

Z490, 10900k (oc). RTX A6000.

Even with maximum video card utilization on, some features barely touch the card at all (2-5%). In the meantime, product will use 80-90% CPU. This happes regardless of parallel processing.

All drivers are at the latest versions.

The WORST case is when upscaling to 4k. For example, have a 1080p video @ 30fps and set it to upscale to 4k@60fps. Even with the RTX A6000, rendering is < 4fps

So I have my GPU passed through in a virtual machine, there’ll be a small performance loss mainly around CPU/RAM performance - however, I’m currently running 3x files in parallel from 720p to 1440p and also doing 30fps to 60fps on them and each one is at around 0.80-0.82spf.

So when you work that out it’s effectively 0.27spf which to me is pretty good for a mid-range GPU. The key here is VRAM.

  • CPU Utilisation: ~68-85%
  • Dedicated GPU Memory Utilisation: 6.8/12.0GB
  • Shared GPU Memory Utilisation: 1.7GB/12.0GB
  • GPU Package Power: 90W
  • GPU Temperature: 50°C / 122°F

I could probably tack on one more 720p file to run 4 in parallel but I usually just queue stuff up so when 1080p is hitting it I would likely run out of VRAM with 4… so 3 is a good choice for me.

Anyway - with 3.x and being able to run in parallel I have no real complaints. Your opinion and usage may vary - but please let’s have open discussion about it as it may help the developers in understanding use cases.

I find TVAI quite fast compared to open source upscaling offerings through things like REAL-ESRGAN where it is many seconds per frame .

With an A6000 and the VRAM it has, have you considered splitting your input files and running the upscales in parallel?

Which CPU and RAM (speed is also relevant) do u have ? Without fast CPU and RAM you won’t be able to utilize the 3090 properly.

Also, close to full utilization of a 3090 and above GPU can usually (depends on model and resolutions) can only be achieved by running at least 2 tasks in parallel.

Does parallel mean opening the program for each video?
Or running 2 video files running in same program?

In v2.6 u need to run the program 2 times in parallel, because it can only process 1 video at a time.
In v3 u can process up to 4 videos in parallel with one instance of the program