I'm not familiar with Topaz Labs and what upscaling looks like in terms of CPU/GPU loading. Is it taking an hour+ of loading all the cores (and or GPU) to get the auto fan setting to spin up into audible whine territory?
My use is more data science, so I tend to load (usually CPU) cores for a few minutes and then they're relatively idle as I'm writing code. But, I'm running a lot of containers these days, so a bunch of cores and RAM would let me give Docker a lot more resources, and basically never have to think about what's running (i.e. manage resources).
It's fairly intense video upscaling, it uses a lot of the GPU and ANE.
I also run docker containers, which apart from build, will tend not to have much effect on CPU. I assume you're training externally? If that's the case, your workload looks to be fairly intermittent and thus the fans will rarely spin up.