Performed more stability testing. 4.2GHz (40 x 105MHz) @ 1.25V with the memory at 2520 using 14-14-14 timings @ 1.20V. High LLC was also used. Room was at 75° F. The CPU temps will never get this high during normal use, so I think it did pretty good. I was also browsing the web during the earlier parts of the test which caused the GFlops to go down a bit. Went from 162 Gflops at stock to 228 Gflops using the same LinX AVX settings of 14 threads and 24GB of memory. A nice increase. Going to do long duration testing before I call it stable.
I'd like to point out several errors in Your testing:
1) You are using wrong number of threads - Linpack is FP type load that does not scale with HT at all. By running 14 threads you are at mercy of Windows scheduler to swap them around. The proper way to do it is to run Linx, set threads to CPU physical cores AND in task manager, set Linx affinity to physical CPUs only (basically all odd or all even selected).
2) Linx 0.6.4 is ancient, simply put you are getting less GFlops than Haswell quad on 4.5Ghz and thermal load is not full either. Getting Linx with updated Linpack libraries ( latest is LinX 0.6.5-11.2.0 ) is great idea. ( tho you still need to apply threading affinity from (1).
3) I would not recommend starting testing AVX2 on high voltage / freq, drop some and start from there. Also using tools like PerfMonitor 2 is highly suggested, that way you can see unhalted CPU cycles from each core and if they drop during main Linx calcs phase - you have throttling.
P.S. AVX2 testing is as real world as any other stress test out there, same caveats apply.