People claiming to be stable with "amazing" results should be doing 1 hour of Prime, 768–900K in place FFT and posting a screenshot.
They also need to post their room's temperature and test inside an actual computer case.
I just hooked up my lapped 8370E system and my results fall right in line with what Anandtech's 8320E review found — and this is with a custom loop (480 with four slow sleeve fans, 360 with Vardars), a 200mm intake on high, a 140mm fan blowing onto VRM and out the back, no cover for the back panel, a 140mm fan blowing onto the northbridge and out, a 1200W gold PSU, and a cheap board (Gigabyte UD3P 2.0).
CPU Temp: max 52.8, average 48.1, min 19.3
NB Temp: max 45, average 44.3, min 33
Vcore: average 1.449, max 1.452, min 1.428. BIOS set to 1.435.
NB Core: average 1.272
VRM: average 86.2, max 89, min 48
Power (POUT): average 174.5W, max 181, min 52.3
I got hardware errors in Prime when BIOS was set to 1.431 so 1.435 seems to be the cut off.
The VRM temp exceeds the spec at load as far as I recall from The Stilt's data, which isn't great but I don't think it's hot enough to throttle. I thought this board was throttling over 4.4 but I think it's a bug in LinX. Unfortunately, there are no newer versions of that available with the AMD patch.
Anandtech's 8320E overclocking test used inferior cooling but an open test bed. Their board had more droop so they needed a much higher BIOS setting (probably didn't use the optimal LLC setting but it is AsRock so...) but the load Vcore is right in line with what I've gotten.
Guess what I got with my 1.449 average VCore? Just 4.7 GHz.
4.7 GHz
2354 NB Clock
2568 HT Clock
214 bus (necessary on UD3P to increase BCLK to deal with multiplier boot bug)
I frankly don't believe the magical results. Maybe it's my motherboard being weak but given that my results are basically the same as Anandtech's with different gear and I've also tested an 8320E (including on air) I don't see tons of silicon lottery involved.
Now, testing without a case and chilled water I'm sure I could do a bit better. But, that's not real world.
*UPDATE* Despite passing an hour of in place 768 – 900K FFT I could not stabilize 4.7 with running large FFTs out of place with 13.5 GB of RAM. Core 5 failed 12–28 minutes in, routinely. Adjusting NB Core voltage, RAM voltage and speed, and VCore didn't help. Given the way the power consumption figures in HWINFO seem to max around 180 watts for input I wonder if there is some sort of odd VRM throttling that somehow allows core load to still read 100%.