Quote:
Originally Posted by
Asmodean 
Not meaning the hijack the thread, but seeing as TwoCables is here
I know it's the norm to test for 12 hours to let a full cycle of FFTs complete after an overclock. What I was wondering is what the
real point of doing so is? not questioning the method of course. Hypothetically speaking, if your overclock was not stable, does it cause crashes, or would it be counter-productive, and harm performance also?
I was just always curious, if an unstable OC just caused it to crash, or also lower performance. Not a blatantly unstable one ofc, but not completely stable, I mean.
I've heard that instability can result in corrupt files or data, but I don't know yet if there's any truth to that.
I've also been told that the more stable an overclocked system is, the easier the overclock is on the CPU. That's not to say that I assume perfect stability results in the same lifespan as never overclocking, but I imagine that it means it allows a longer lifespan than otherwise might be possible with instability.
I think instability affects performance too, but it may only show up in benchmarks. I'm not sure.
For me though, it's peace of mind that I can relax knowing that I'll never see a lock-up or a BSOD, or any problems whatsoever. That's what I've been enjoying ever since July 11th of 2011 when I upgraded to Sandy Bridge and overclocked it to 4.7 GHz. I used Prime95 v26.6 build 3, but I don't know how many FFT sizes it has in the Blend test. Although, my system is stable enough to run the Custom Blend test in that version of Prime95 for a bit over 12 hours with about 95% of my memory utilized (as opposed to Prime95's default of 1024 MB).
Since that time, a new version of Prime95 came out (version 27.7 build 2), and I found out that it has 82 FFT sizes that the Blend test goes through. I intentionally found this out by performing the Custom Blend test with it set to 1 minute per FFT size. Instead of taking 82 minutes, it took 105 minutes! So if someone wants Prime95 v27.7 build 2's Custom Blend test to go through all 82 FFT sizes at 15 minutes per FFT (which is likely much more stressful than 1 minute per FFT), then they would probably have to have a goal of about 24 hours! I mean, 82x15 is 1,230 minutes, and 1,230 divided by 60 is 20.5 hours. I mean, at only 1 minute per FFT, it took 105 minutes instead of 82, so I can only imagine how much longer it would take than 20.5 hours at 15 minutes per FFT.
Now, I suppose a compromise could be reached by doing 10 minutes per FFT size which would be exactly 13.67 hours before considering the extra time in between each FFT which could result in the final time being about 15-16 hours I guess.
Although, I have to consider the fact that I didn't know this stuff back when I tested the stability of my 4.7 GHz overclock in Prime95 v26.6 build 3. All I knew was "12 hours", so that's what I did and it's been perfectly stable ever since. It's so stable that it's almost boring! :)
Edit: I say linpack is useless if you have a CPU that has the AVX instruction set (such as Sandy Bridge and Ivy Bridge) unless you are lucky enough to have downloaded one of the linpack programs that take advantage of AVX. If not, then your linpack tester is not stressing the entire CPU and is probably a waste of time. Prime95 v27.7 build 2 uses the AVX instruction set, so I assume that this makes it the best stability testing program available right now.
Edited by TwoCables - 8/16/12 at 5:16am
