Here it is:
Board: R5E10, cooling through a monoblock
CPU: 6900k, cooling through the monoblock
RAM: 64GB, 3200MHz
GPUs: Nvidia GeForce GTX 1080 Ti (2x), original cooling
SSD: 960 Pro
PSU: Seasonic, 1250W
I run the RAMs at their XMP settings, but manually entered the values
I had a stable OC on CPU at 4.4GHz, cache 3.4 - was passing every test, including the RB; then reduced to 4.0 GHz, cache 3.1GHz (RAM remained)...
I did memtest86+ and HCI memtest 500% in both settings with 0 errors... Also the GPUs are stock...
My OS is WIn 10 Pro. Fairly new... killed Cortana and uninstalled Intern Turbo Boost 3.0 (this showed up after and auto update). Don't know yet how to stop the auto-updates.
Initially I had 378.92 driver with GeForce Experience (no problem observed that time), then the thing auto-updated to the 381.65 - when I observed trouble... I used the driver removal tool (DDU, I guess is called) in safe mode, then clean installed the 381.65 only (no GeForce experience)... problem is still there...
3DMark Fire Strike works as a charm, can kill it only if I push the GPUs too high.
RealBench 2.54 testing runs as previously - does not show any problems in 2 consecutive runs.
I get the problem (Luxmark thing) when stress-testing with RB around 10-15 minutes... temps are fine (mid 50s at 4.0GHz)...
Oh, and I have GPU Tweak II and AIDA64 running in the background - did not try to kill them yet...
Let me know if I forgot something...