Technical comment about GPU load while processing:
As it is quite difficult to encapsulate all kind of load, stressed to GPU computational structures, into one value, Microsoft used a simplification.
This idea is somewhat good because it is independent of various processors internal core units (like CUDAA, Tensor, etc). Notice, that almost every graphic card consists of a different number of such units, with various bus speed, and inter-bridges speed (frequency and data width).
Computational speed varies also by throttling coming from thermal or power limitations.
Making long explanation short: the percentage of a load of GPU presented by the Task Manager is a proportion between actual memory bus load (actual data transmission) to maximum possible data channel capacity. Simple enough to be applied within the same regime to all types of GPUs. It is more accurate while gaming (where data transmission is more proportional to computational load) and less accurate while strict data processing (computing).
Real load of GPU during Topaz Video Enhance AI processing relies more on computational power, so the Task Manager will always show lower GPU load (especially compared to more specialized measurement, like GPU-Z load calculation).
Newer versions of Topaz VEAI show somehow lower computational load at the Task Manager. And this is not bad. This comes from the better internal load (more optimized algo, with less VRAM-to-Core transmission).
Always use reliable measurement - GPU-Z is much more accurate.
Usually, for a VEAI task, a real GPU load will be 50-100% higher as presented in Task Manager (lower for UHD resolutions due to more memory transfer demanding operations). Verify yourself at GPU-Z 
4 Likes
