What magic mask tracking utilize CPU or GPU?

I have DR Studio 18.b4 and I use Magic Mask. When I track it first it does not load CPU fully. Like only 20%. When i render video or create proxy it load all cores 100%. Perhaps it is smart not to load CPU 100% on Magic Mask tracking but 20% is too low in my opinion.

Another thing is that GPU is not utilized at all. 0%. Although on some topics I read that people refer to better GPU for faster process.

SO my CPU at 25% load and GPU at 0% load. There is room to speed it up. How could I force Magic Mask tracking to use more power?

Magic mask is gpu based as it uses a machine learning model for image segmentation. Gpu compute usage does not properly show in task manager.

Not sure if Resolve has a cpu fallback mechanism for its ML tools, I don’t think it does.

That is why it is so slow. Perhaps I have old GPU that does not very effective with AI.

Perhaps, but for anyone to actually tell you should specify your system specs and OS.

There is a special form for version 18, Sergey.

viewforum.php?f=36

You need an Nvidia GPU with tensor cores (RTX 20 or newer)

Or an Apple M1.

How did you measure the utilization?
If on windows, you can't trust the default task manager settings.
If on windows, use gpu-z and it's sensors instead, it's a dedicated program to show exactly how much of the gpu is used and what type of usage.

On my computer magic mask uses 100% of the GPU compute resources, so something's either wrong on your end or wrong in the measurements.
panos_mts is wrong as well, you don't need tensor cores (ref my 1060 does not have any, yet magic mask works). In theory they make the calculations faster, but not required.

Second the use of GPU-Z, a much better way of seeing what your GPU is doing.