Originally Posted by Scorpion49
What settings are you using for that OC? Something is definitely wrong.
After reapplying the thermal compound again (it looked like I actually had too little this time), I was able to get it up to 4.7 GHz while keeping the cores at about 61c max (usually hangs out at about 58c the majority of the time during stress testing) and socket around 65c. Not bad, but the voltage is pretty high for 4.7 giggles because I didn't get lucky with this chip - it takes 1.548v (including LLC) to stabilize it. This is with all 8 cores enabled and all power saving features off (C'n'Q, Turbo, unused core disabler, C1E, APM, SVM, spread spectrums, etc.)
Looks like it has a stock VID of 1.380v (ouch), although it'll run all the way up to 4.4 GHz stable on 1.35v, which is odd. Not sure if ASUS is just misreporting the stock VID or what. It also had my CPU/NB voltage way high, I think it was 1.35v, so I took it down to 1.175v to get rid of some extra heat there as well.
Seeing how it's not recommended to take this chip above 1.55v, it looks like 4.7 is the max OC on this one. Sometimes you get lucky, most of the time you get an average chip, and other times you get unlucky. It looks like I got unlucky. Stock VID of 1.380v though?? It almost seems like it should have been binned as an 8320 instead. It must have BARELY made the threshold for the 8350 bin. I think the H220 probably could have pushed it further with lower temps if it was a decent chip and could go nuts with lower voltages, but such is not the case with this one. Unless my Prime95 errors were memory-related, I'm not sure. The memory is rated at 1866 with 9-10-9-28-2T @ 1.50v, which is exactly where I have everything set. Bad memory module? Maybe. It's very difficult to distinguish where P95 errors come from.
It still errors in P95 @ 4.7 with 1.548v, which has GOT to be pretty much impossible for the CPU. I've heard rumors that P95 hates the FX-83XX series though. It passes IBT with 20 iterations and has run stable in Windows all day today, including a bunch of benchmarking going on with Unigine, Sandra and 3DMark11. I feel like I might have to resort to the "use it and see if it ever bluescreens" method of determining stability, as much as I hate to do that. I'm tempted to Memtest it, but I've seen Memtest go for 24 hours without finding anything when the issue does turn out to be memory.