Had to chime in here, and I can see my post could easily become another thread topic.
"First time OC'ing . . . " Welcome . . . to our obsessive-compulsive addiction, AzNKiD!!
I started doing "it" around 2004. There had been a time when I build a new PC for myself every year or two. Now, my pattern has changed. I would think that the more you do "it," the greater your accumulation of experience with "it." But there are caveats.
Every new processor generation means a new chipset generation, and every new chipset generation seems to add more extra features for overclocking. Take for instance the bCLK "strap" available with IB-K processors but unavailable to SB-K -- for the Z77 chipset. [Do I have that wrong? Say sumpin!] Then there is the addition of cache speed and voltage, which wasn't available on boards a few years back. So with every processor and chipset generation, there is a learning curve.
IDontCare -- one of our truly renowned contributors -- offered some insight into finding a "sweet spot" of voltage for an overclock speed. He suggested that there is a voltage range, just above "unstable," in which processor errors occur but get corrected. The focus of his attention was the LinX stress-test. Running it with "affinity" set to four threads and one for each core, the program generates consistent GFLOPS results for each test iteration. IDC suggested that noticeable variability in these results may be symptomatic of the errors that occur. In other words, a list of 20 or 30 GFLOPS results with a higher range, standard error or just "variability" may suggest a voltage that is at a margin for being too low.
With that, I've sort of changed my overclocking strategy in hopes of minimizing BSODs during stress-tests: I deliberately start testing a clock setting with a voltage that may be a bit too high, and then I run samples -- 10 or 15 iterations -- in affinitized LinX to watch the GFLOPS results.
Other than that, there is a "race" situation once described by ShintaiDK when I asked why stress-test programs didn't trap the errors or failures, instead of generating a BSOD.
But even -- I could even say especially -- the BSODs are symptomatic for choosing the next step. The BSOD stop codes like 101, 09C, 124 etc. indicate what feature may be causing the crash. 09C or 124? Consider raising the IMC voltage. 101? Bump up the VCORE.
There are also different opinions about the length and type of stress-test which guarantees 24/7 stability. The traditional view suggests one needs to run Prime95 for 24 hours. Keep in mind that there is an "sFFT," "lFFT" and "Blend" test in Prime95. Further, errors not shown in the intensive sFFT "CPU-intensive" test may show up in lFFT if there is a problem with the RAM configuration.
Prime95 still has useful value. On the other hand, if we can reduce the length of these tests, it would speed up the tuning and fine-tuning.
OCCT provides a "CPU:OCCT" test and a "CPU:Linpack" test. The author argues that his CPU:OCCT test will find errors early -- within a 3.5 hour test run. It also finds those errors without generating the same sort of temperatures you find with CPU:LinPack, affinitized LinX, IntelBurnTest.
I was able to convince myself through test results that his claim is reasonably true. So I run CPU:OCCT first of all -- for 3.5 hours. Then I run the LinPack version for 4 hours. I move on to LinX for a prelim run of 20 iterations, followed by the "samples" I mentioned. Once all that is done, I may run either LinX or IBT for 30 to 50 iterations to "validate" stable settings.
I also run Prime95 sFFT and lFFT for 4 hours each.
If you're seeking to overclock a similar processor, same socket, then the second time around it should take much less time.
But finding the very lowest voltage at which you think you found "stability" isn't likely enough. You should keep notes on your BIOS changes, the length of successful and unsuccessful stress-tests alike -- and that includes these "threshold" voltages. Once you've found the threshold, you want to bump up the voltage settings a couple notches and start looking at GFLOPS results.
And what I should've written at the very beginning: test the whole enchilada at stock settings first. In fact, before you even install windows, you should test the RAM at its stock settings thoroughly.
In the midst of all this new discussion on stress-testing, I find the latest advice to turn off the AVX2 parts of these tests, to avoid temperatures that would never occur in any meaningful real-world usage of that part of the instruction set. I suppose that makes sense.
On the matter of temperatures. Which -- also raises a discussion about cooling strategy. The processor is supposed to operate in a range of temperatures at stock settings. Temperature increases by the square of voltage increases, and linearly with clock speed. I usually aim for the TCASE spec with my cooling strategy, although there's a prevalent view that you can heat these puppies up to the point where they throttle at the TJunction temperature spec. There IS NO TCASE sensor, and you would expect core temperatures, average peak core temperatures or package temperature to exceed the mythical TCASE.
So -- two thoughts. First, there is some limit for reasonable voltage settings, but it has become less evident what it might be since Nehalem, when Intel last published a "maximum safe" voltage spec (for a 32nm processor). Second, you'd like your overclock configuration to operate in a temperature range similar to Intel's stock intentions. Third, you are not likely to see lower temperatures whereby you can achieve the same overclock at some noticeably lower voltage, so it is possible to "over-invest" in cooling. But if the processor operates at temperatures that are very high, there may be increased electrical noise generating instability, and the temptation to keep raising voltages which only increase temperatures.
And there's nothing wrong with OC'ing these processors while leaving EIST, C1E and the other energy-saving features turned on. What you will find by using a "fixed VCORE" overclocking strategy, should be no harder to find with those features and "turbo" enabled.
Finally, this is an obsession. You begin to think that higher Ghz is better, but benchtests may prove otherwise. Overclocking becomes an end in itself. It's probably best to keep this in mind as you go through the process and make decisions as to whether you found a "good clock" and choose to stop pushing it further.
Facebook
Twitter