I wish the program didn't increase the wait time for checking in with the server again, one my machines ended up on a 2 day wait, rebooted and immediately got a CPU WU.
From my understanding, the FahCore works using OpenCL, so if those Tensor and RT cores can process anything dealing with OpenCL or give a boost it would make perfect sense why an RTX card rapes when folding.
The OpenCL thing is why most older cards will never get a WU, though they could in the past. It's more of a "Universal" thing so the FAH team isn't having to code for specific GPU's anymore.
EDIT: From my 5 second google search, it looks like Tensor cores will dramatically boost some OpenCL tasks. Would explain why it takes my poor 1650 around 5-6 hours for a single WU.