I now solved my issue with all WUs failing with CUDA error 999.
The error I was receiving was the following:

ACEMD failed:
    Error initializing CUDA: CUDA_ERROR_UNKNOWN (999) at /opt/conda/conda-bld/openmm_1589507810497/work/platforms/cuda/src/CudaContext.cpp:148

After looking around, I found this thread, where the error 999 was encountered after entering the sleep mode. Simple reboot fixed my issue. 🤦

I crunch only during office hours*. It looks like my GTX 1060 will just make the 5 day limit on these biggies, but only if I hand massage it each week. :( If I start one each Monday then the weekend won't get in the way.

(* more like 10-12 hours, computers are set to preheat my little office before I arrive.)

No, normal WU shows me a % after 1 minute after star.
It`s took them aabout 5 minutes for 1%.
Big one would show 0,333 after a hour.

No, normal WU shows me a % after 1 minute after start.

The present workunits are exceptionally long.

It`s took them about 5 minutes for 1%.
Big one would show 0,333 after a hour.

It takes 4 minutes for an RTX 2080Ti to reach 0,666% with these workunits.
Your GTX 650 is too slow for these workunits, you should set GPUGrid to "no new tasks" until this batch is over.