I am getting the same issue as you.
Usually RB reports instability after 3 goods passes (and fails at ~25 minutes)
with the same Blender:BLF_lang_init: 'locale' data path for translations not found,
but sometimes can run for a longer period. 16GB test seems to be the most repeatable at failing,
8GB seems to be more solid.
It has driven me nuts for the passed few weeks, and I am clueless what to try next and where might the problem be.
Setup: Win10, Z170-A (R1 board, bios 3007), 6700K, Crucial 32GB/2666, STRIX1070-OC (default gaming mode), HX750i
simple cooler master air cooling for cpu, Fractal R5 case with plenty of air flow, 4.0/4.2GHz
NVidia driver 376.33, 1070 micron bios updated, and all other updates should be up to date
Tried with different ram configs, 16 ,24, 32 GB and all speeds, 2133, 2400, 2500, 2600, 2666
Underclocked CPU to 3.5 GHz, gave plenty of power 1.35V to VCore, added power to RAM: up to 1.220V
Under most test CPU core temps upto 80C, gtx at 70C, ram at 52C
LuxMark can stress test for hours on it's own, iGPU and GTX together
CPU only stress test fails after first run, but it is broken imho, since it fails on all system I tried.
Other possible problems that can relate to GTX problems
Can play games for a very long time, no issues, but TC Division has crashed on me 3 times.
Complete system freeze, twice after 4 hours, and once after 13 hours
One more similar freeze happened during MPCHC video playback minutes after starting.
Otherwise no other major issues, can run 24/7 for weeks without rebooting.
That includes running games, stress tests/benchmarks (almost all tried by now), video playback, encoding, what ever.
And finally a small on the side question: Is it normal for shadow effects to flicker in some games/tests?
Looks like they get rendered every other frame. Not complete shadows, just small areas near the edges.