I’m hoping to hear from VideoAI users here and their experience with running on M1 Max or Ultra setups.
I’m in the market for a new home & work machine and I’m looking to see if I can have put together a setup that allows me to run a video upscale in Topaz while concurrently doing work stuff. For work, I usually have browsers windows, office (excel mainly) and then a handful of other software dev tools running.
When folks use VideoAI on an M1 Max or Ultra machine, does it essentially turn into a single-use/purpose device or can I perhaps concurrently use it as a work machine? Or is the only way to limit the resource usage in a VideoAI setting?
Appreciate any insight anyone is able to share!
I have a Mac Studio M1 Max and from your description of other tasks you’ll be doing, I don’t think TVAI will interfere. Even if it might do with default TVAI settings, there are ways within the UI and CLI to reduce the resource use. E.g. Low Power mode within the UI (or in CLI set “instances=0”), or via the CLI you could limit ffmpeg to just 1 thread.
Just to add, regarding performance, TVAI on Apple silicon is currently woefully inefficient when upscaling SD to HD (inc. v3.2.0). I don’t know if this is the “fault” of Apple, Topaz, or both. Upscaling 1080 to 4K is fine with very respectable performance on Apple silicon.
Appreciate you sharing your experience here! Has the performance of SD → HD always been this way or a regression in the latest version?
On Apple silicon, TVAI SD upscaling has always been significantly less efficient than with HD. However, the work-around solution of using multiple (parallel) processes has become less effective with later versions. For example, with version 3.0.1, I could get just over 25fps in total over 6 processes of SD upscaling. Now, with v3.2 (and v3.1), I can get only get a maximum of 20fps. Also, there is now no additional gain in running more than two processes in parallel (at least on my Mac Studio). Even then, there is some faffing around to do to get that speed - run one process on the CPU and the other on the GPU. Oh, in case you’re thinking you can do that by using the “AI Processor” selection in TVAI - WRONG! That is ignored on Apple silicon Macs. The effective way of getting the process to run on the GPU cores is to restrict the max. memory (eg. 10%). If the max. memory is left at 90 or 100%, then SD upscaling will run on the CPU cores.
Having said all that, I’ve now found a way to upscale SD much more efficiently, getting over 26fps. I’ll write about that separately. For now, I’ll just say that it involves multiple stages of ffmpeg processing and makes use of segmenting and stacking, to increase the TVAI input resolution for maximum efficiency. To fully automate things, I’m currently creating scripts, which I’ll make available.