Yes, if Luxmark freezes, but everything else in RealBench works, you're still on the edge of instability somewhere.

Luxmark especially pushes I/O transfers from the GPU though CPU and RAM. It will work the memory controller hard.

If you set your RAM to stock voltage and speed and it still locks up, then look closely at your CPU voltages. SA voltage comes into play here, as does the Vdroop effect of CPU load-line calibration.

I have two near-identical i7-6700K builds. (Only difference is the video cards.) The one that's "mine" (the one in my sig) seems to be a particularly good i7; it overclocked easily to 4.9GHZ at temperatures I liked. (5 was possible, but too hot for my taste) Luxmark never failed on it -- Instability showed up more on Handbrake tests that used AVX and pushed CPU loads to the max. The other build, which I put together for my wife, was an uphill battle all the way. The memory controller showed a distinct preference for one pair of memory slots over another. Luxmark would throw it constantly until the SA and VCCIO voltages were tweaked slightly beyond just what was necessary for good CPU clock stability. My impression is that one's memory controller is "weak" and needed additional tweaking to make it run well with the OC'd RAM. The end result was that the best stability came from a slightly lower CPU overclock (4.7 max instead of 4.9) which gave a better range of voltages to keep the memory controller happy. The silicon lottery is like that sometimes.

I was chasing my tail for days trying to setup a 6800k overclock, getting black screens and luxmark crashing just like in your screenshot, turns out it is indeed a gpu issue there is an open issue with luxmark crashing on pascal gpus as listed in the release notes for even the newest nvidia driver, to verify it is a gpu issue you can select benchmark instead of stress test and only select h264 video encoding, or you can run the x264 stress test people have been using on haswell chips for years or run realbench stress test with a non pascal gpu. Until nvidia fix this issue do not use realbench stress test as it is not reliable on pascal gpu based systems and you will go crazy.. Hope that helps!

Very often a clean (use DDU) reinstall of graphics driver solves this issue.

Been posted up here many times if you have a quick search.

There are some issues with Pascal and latest drivers though.

Hope Nvidia get their act together RE drivers ASAP...:(

Running the latest NVIDIA drivers and benchmarking in RB without a problem.