Inconsistent Results between IntelBurnTest 2.54 and LinX 0.6.4 with i5 3570K

[Loser777]

I'm currently running a i5 3570K on an MSI Z77IA-E53.

A few weeks ago, I got this result while stress testing on stock voltage.


While keeping things as stock, my system crashed yesterday, so I decided to retest things.
Now, LinX 0.6.4 would fail 2-3 iterations in @ stock voltage, 4.2GHz.
This behavior agrees with what IBT 2.54 found, but it took more iterations:


I dropped the multiplier to 40x to do some more testing there:
Now, IBT 2.54 shows no issues with 10 iterations at maximum stress (usually considered to be a decent indicator of stability), while LinX 0.6.4 again fails after a few iterations:
IBT:

LinX:


The interesting thing is that IBT uses AVX (greater number of GFLOPS) and also produces higher temperatures in RealTemp, so I would expect it to fail first, yet LinX with its slower execution time is failing first.

Above all, I don't understand how something that was stable for three hours of LinX begins to fail so suddenly--I may try a CMOS reset and see if that gets me anywhere.

[Loser777]

I noticed an interesting effect:
I dropped my memory speed to 1066 instead of 1333 (timings changed as well based on SPD), and now it seems to be more stable (currently already doing > 5 iterations) @ 4.2GHz, stock voltage. I should be able to confirm if 4.2GHz is stable in a few hours.

Well, LinX pasted 25 iterations (~15GB/16GB used) without AVX... and failed about two iterations in LinX with an updated binary. I'll try and see if I can get 4.0GHz stable now (this is quickly becoming really pathetic).