Based on what the Topaz developers have told me, the best use of multiple GPUs to speed up exports is to run two instances of TVAI, each one using a different GPU:
- Configure TVAI to enhance the first half of video using one GPU.
- Open the same video on another instance, configure to enhance the second half using the other GPU
- Export both simultaneously.
- Splice the two exports together.
I tried this, and it definitely works. Maxed out two GPUs with 18% CPU utilization. My question, is, could this process be automated? A “divide and render” button in the UI, or maybe a script?
I did try that. Two instances on the same GPU reduces the fps by 1/3 to 1/2. Would probably work better if the GPU wasn’t already hitting 100% with just one instance.
The same principle could apply on Apple M1/2/3 in situations where the Neural Engine can be used (e.g. upscaling SD using Artemis or Proteus). You can treat the Neural Engine as a separate GPU. So it’s possible to get higher overall speed by upscaling one video (or segment) using the GPU cores while upscaling another using the Neural Engine. I totally agree it would be great if TVAI could do this automatically. Thanks.
