Are you also stress testing for stability?
I got stress test failures and/or crashes (though normal use seemed fine) with a combination of undervolting, high ambient temps, and memory at 3200MT/s. I don’t know what was the cause, so for now I’m back at stock (auto) voltage.
I am stress testing with OCCT and thus far (-0.5mV) I haven encountered any stability issues.
How did voltage vary in actuality? Curious to know if it was still drawing the same at the wall or if it changed significantly.
Min/Max are decreasing, average is roughly the same and not (really) significant, since I measure over a few days with a varying CPU workload.
What I tend to do to test 100% stability is to run handbrake hevc with multiple bluray files and eventually it freezes. I have never been able to actually get it to rune stable without freezing in handbrake. I could have prime95 and aida64 stress test running for an entire day and no freeze, but handbrake is the one that ends up freezing it.
As mentioned above, I don’t try to achieve a 24/7 undervolt. Which is why OCCT seems sufficient for my purposes. However, are there any sample bluray files to download? I’d test my system with handbrake as well if I had any of those.
Why not?
Because it is often harder to test for 24/7 stability (and it takes quite longer), than getting it to run in most cases. (However, I keep my undervolting in mind in case some kind of error or BSOD occurs).
That’s for sure. Quite frustrating when you just want to use your system. But I guess stability is a different level of priority for different folks.
I also dislike letting it run and doing a stress test in the summer and for some weird reason, I decide to overclock/undervolt in summer 
My old 2500k got to 4.3GHz without any problems too. (I also didn’t 24/7 test it). A friend of mine’s PC crashes during Prime95, but is stable when gaming/workin (i5 4560k (something like that).
BTW - I’ve now concluded my -50mV tests with the following results (13 CPU-Z runs):
-50mV
Single: 476.12
Multi: 4804.9
I decided to try Cinebench with different undervolts and thus far there seems to be a performance increase. -0.25mV is 0.8% (multi) faster and -50mV is 1.7%/1.3% faster (single/multi). With stock and -50mV settings I’ve already done 9 tests over three days, with -25mV I’m at three. I’ll keep this thread updated, but wanted to share my preliminary results with you guys.
To be honest, this seems to be more in line with the way SenseMI works than my CPU-Z results. Furthermore, in Cinebench my CPU clocks to its max in the single threaded test, in CPU-Z it remains at all-core high.
(I updated and concluded the initial post again)
You win the CPU lottery my friend!
It seems so. Although, I’m not sure how other 2700X perform in that regard…