Originally Posted by HornetMaX
Originally Posted by shad0wfax
The 140% current, extreme phase control, and manual VRM at 350 should get you all the way to 5.0 GHz, assuming the rest of your system can handle that frequency, should you ever decide to push that far.
I also use the Duty Control Extreme and recommend that as well.
LLC Medium is a value that I came up with on my own.
Spread Spectrum disabled is also something that you have to experiment with on a case-by-case basis.
I've duplicated most of your profile (as seen here
), except the manual CPU PLL voltage (left it on auto).
At x45, offset +0.050V and +0.030V pass a quick 1344/1792 prime95, while offset +0.010V crashes after a few seconds.
I let a custom blend run the night (x45, offset +0.030V) but it failed somewhere.Noob inside:
I didn't set realtemp to log to a file, so I don't know when it failed, as I've found the PC restarted, already in Windows. Anyway, it failed after more than 3 hours. Maybe I can see time (and failure code) via the memory dumps done by Windows.
I've just started another one at +0.040V. Starting to think that the Auto-OC was not that bad
Ah, I've changed my profile to a more efficient setup. Here's the link to my new set-up:
The PLL voltage is usually not necessary to tweak, but I actually reduced mine clear down to 1.550V in the above link and had good results with it. I've actually booted into the OS at 5.2 GHz now, can run 5.0 GHz stable (but with 89-92C temperatures in IBT / Prime Small FFT that I'm not at all comfortable with) and I'm working on an efficient and low-temp 4.9 GHz profile, if I can get it 18 hour Prime95 custom blend stable.
Do check my new settings though for a more efficient model.
Also, you can check to see where your tests failed by consulting the Prime95 folder and looking for the file "results.txt" where you will find the last successful FFT lengths tested.
Then you can check this list of all FFT lengths that a Prime 95 Blend test will run on an i5:
This is a list of all FFT lengths that will be tested on an i5 in Prime95 Blend (Click to show)
Most Sandy Bridge seem to have trouble with the 1344 and 1792 lengths, which are tested in that time period. However, on my machine, it's the test #60 at 2688K that will cause me to fail usually, and that doesn't happen until around the 15 hour mark at the default 15 minute time frame.
If you're going to leave the default time to run each FFT at 15 minutes, it will take you about 17.5 to 18 hours to finish all FFT lengths in a Blend test (assuming that there aren't core imbalances.)
If you shorten the time to 10 minutes per FFT length, you can test all of them in about 12 hours, again assuming no core imbalances.
If you're looking for a quick and dirty OC test, just to see if your latest setting is moderately stable, you can shorten the time to test each FFT to 1 minute and after around 70 minutes you should be done. :)