Some guy said that logic dictates that if only 1 application out of 5 fails consistently then there must be something wrong with the program.
Actually, that is not the case.
When a software engineer makes an application such as this, she/he must decide what to test and how to test it.
For all you know, most of the "test" applications you use don't actually test anything. That is, it could very well be that they just give the CPU some work to do and if that work finishes then you "passed". You have no idea if the computational results of each operation were even checked by the software.
Simplified example to show the point:
Think of these as series of operations in 1 run:
2 + 5 (= 7 answer ignored, not verified)
5 + 8 (= 20 answer ignored, not verified)
23 / 3 (= 7.5784 answer ignored, not verified)
45 / 5 (= 8.9999 answer ignored, not verified)
Congratulations! You passed.
(Because each operation completed without the app crashing)
A smarter test would instead verify the result of each operation:
2 + 5 = 7 (true)
5 + 8 = 20 (Wrong, programmer bothered to check the result, now must decide to terminate test or let it run)
23 / 3 = 7.5784 (Wrong, programmer bothered to check the result, now must decide to terminate test or let it run)
46 / 5 = 9.9999 (Wrong, programmer bothered to check the result, now must decide to terminate test or let it run)
Keeping this simple, as you can see, you have no idea what is actually done by these applications at even such a trivial level. Even if some operations are checked other operations may not be.
Adding to that, you don't know what actual operations are being performed. In the end, at the assembly level, it is all simple registry operations, which if you know anything about programming would know that these operations are CPU agnostic across x86 x86-64. In other words, they don't have to rewrite the app just because AMD release a new x86-64 CPU...
Prime95 does not have FX8320 issues because I have ran it for 12hrs with no errors.
Overclocked, for me, Prime95 stability is vcore related. With less vcore test fails sooner. So it is obvious that this is an overclock issue and not faulty code issue. The code doesn't become better code proportional to the amount of extra volts I add (test runs longer)... The code is fixed. With high volts the overclock is stable because the results of the operations performed by prime95 are correct and if I give it less volts the results are not correct. Simple as that.
I have read what the IBT guy said (that IBT can heat better and find stability issues faster than Prime95). In practice I don't find this to be true at all. It doesn't heat up my FX8320 or 990FXA mobo any better than Prime95. Nor does it ever find errors for me. Even when extremely unstable and Prime95 finds error in 3 min. Most other benchmarks barely heat up my FX8320. I've ran all the top of benchmarking apps and my scores are what they should be but these apps barely budge the stock temps. Prime95 maxes them (at 60 deg C).Edited by bios0110 - 11/30/12 at 10:33pm