Even if AMD published the specs you seem really carefree regarding checking your CPU's temps while stress testing. 8xxx series are capable of taking some serious abuse, just a glance over Xtremesystem forums can prove you that. I just don't see how a monumental mess up from your part should be turned into AMD's fault.
Next time just get better cooling or paste, be more conscious regarding checking temps when stress testing or just dont overclock at all
All pertinent details have already been mentioned by me in the "Long Story" portion of the OP. I'm sure you had no patience to read it, but it would be nice if you did before you made some of your comments as your comments are already answered by the details there.
Anyhoo, to recap quickly:
1.) I was not "carefree", as that implies I didn't care and just said "start the test, the hell with you, I'll be back later". I in fact monitored it until it settled (at 74-76C), and I resumed working on my laptop while still sitting in front of the PC being tested, so a quick movement of my head allows me to glance at the progress so far. Everything that happened (except the death) happened on purpose, with the rationale already clearly expressed in the "Long Story" portion of the OP, to recap:
Here's what I was thinking. AMD doesn't publish max temps AND vcore ranges for this part. If I go by the FX-8150 (direct ancestor), max temp should only be 61C (but at what range of operating voltages? AMD doesn't publish). But Trinity mobile SKU's (like the A4-4300m I benched in VC&G) sporting the same Piledriver cores have a much higher limit (sites like CPUBoss and CPU World specify 100C, despite a vcore range of up to 1.35V). So, the same old Piledriver cores can take 1.35V and 100C? Well, if so, I'm still relatively ok (1.396V, 76C).
2.) I don't see a monumental mess-up. If you take the rationale quoted above, given that AMD publishes absolutely no thermal guidelines and operating voltages (unlike for the Phenom II chips), it's as good a guess as any, considering I based the "guesstimate" on a similar Piledriver-core based SKU, and not a Phenom II-era chip (where the common ~62C / 1.45V on air guidelines came from).
3.) I also don't see where I actually blame AMD, or where you get that idea. In the post where I say I could blame AMD by saying "if only they published guidelines for voltages and temps", you'd also see (if you read it again) that I quite clearly and repeatedly state the experience has not aggravated me at all, I am not pissed, and that I accept the death as an inherent risk in overclocking, and even clearly stated in earlier posts that this is way outside of a valid RMA case. I'm not sure how you can take that and then transform it into blaming AMD, when - as far as I can see - everything that I have posted so far has been bile-free, and I've been pretty stoic in all my posts here so far.
4.) As for the "cooling/paste" issue, the reason for that is also in the long story portion of the OP. It was all that was available, I wanted to see what it could do before I replace it with a superior one in the coming weeks. So this stress test isn't actually FOR the 8350 - I already know that it is stable at the setting using the previous paste (Tuniq). Rather, the stress test done is for the TG-2, to see if it can still keep the 8350 stable. Now we know it can't, in no small part due to a 10-15C delta from a top-tier paste like Tuniq.
Anyway, I hope you (and others) don't misinterpret this as an attack against AMD. In an earlier post (a reply to Shintai) I mentioned I have 4 machines. 2 Intel, 2 AMD (it just happened like that, I didn't purposely build 2 each for balance or anything), and one of the Intel machines is a 3930K (other one is a Celeron G550 that's extremely power-efficient, I love it). That I already have a 3930K but still bothered to give AMD a free ~$200 just to upgrade an old 1090T rig to an 8350 should tell you that bringing down AMD is not my intention here.
Rather, it's an interesting data point that I did not want to remain locked up in my own head only. Ever seen an 8350 die before? Neither have I, but now we all have 1 data point, and it died with only 1.396V and 76C load temps (according to HWMonitor) running P95 Large FFT for almost 3 hours.