i7 Overclocking Guides Help

[Chewie_77]

Hey guys, first post here so hello everyone! Anyways, I got my i7 920 D0 and X58A-UDP3 today. This is my first LGA-1336 system so I'm having a bit of trouble getting used to the methodology used to OC this socket.

Anyways, I've been reading several OC guides for this chip and the first step for most guides is called "Isolate the bclock from the CPU". From what I gather, I am supposed to choose a low CPU multi (x12) and put the bclock to my desired end speed (200).

So what do I do beyond this point? Do I test for stability at 2400MHz CPU clock speed, do I adjust voltages? I'm slighty lost.

[famish99]

So lets start out with another question, have you overclocked before?

[Chewie_77]

Yes, I OC'ed an E8400 C0 to 3.6GHz 100% stable a couple years back. That was easier though, all I had to do was increase the FSB and add a little more vCore.

[kill_a_wat]

you may want to list your pc specs in your sig

[radaja]

here you go.
Heres a great 3 step guide to OC your i3,i5,i7 by miahallen.thanks
3 Step Guide to Overclock Your Core i3, i5, or i7 – Updated!

[Chewie_77]

here you go.
Heres a great 3 step guide to OC your i3,i5,i7 by miahallen.thanks
3 Step Guide to Overclock Your Core i3, i5, or i7 – Updated!

That's exactly the one I'm looking at, thanks!

[famish99]

Yes, I OC'ed an E8400 C0 to 3.6GHz 100% stable a couple years back. That was easier though, all I had to do was increase the FSB and add a little more vCore.

That's basically what you do with this heh. Whereas before you sometimes had to increase the VTT, you on the i7 you increase the Uncore/QPI voltage.

The point of isolating the bclk from the core is so that you can make sure your uncore segments are stable before worrying about the core. So run a few tests with your core multiplier low (try to find a memory intensive test like LinX or Prime95 blend, a test that fits in cache won't do you any good). If you're getting errors there it's because your Uncore/QPI voltage or your IOH voltage is too low. Once all those are settled out, just start bumping up your core multiplier and then increase Vcore just like you did before.

[Chewie_77]

As per the guide, when finding your max bclock it states to set the memory multi to the lowest possible. However, I cannot get my system to boot with a mem multi of 6 and a unCore multi of 12 so I used a mem multi of 8 and unCore multi of 16. Is this still going to work alright?

[famish99]

As per the guide, when finding your max bclock it states to set the memory multi to the lowest possible. However, I cannot get my system to boot with a mem multi of 6 and a unCore multi of 12 so I used a mem multi of 8 and unCore multi of 16. Is this still going to work alright?

Just set it so that it matches whatever spec your memory is.

[Chewie_77]

So I decided to try a luck of the draw OC and it's been Prime95 Small FFT stable for about 45 minutes now, it pains me to know there are still roughly 23 hours to go before I know it's 100% stable.

Anyways, CPU-Z is reporting a vCore of 1.280 and my VTT is set at 1.375 in the BIOS. Should this stress test fail, should I give it more vCore or more VTT? My temps are hovering around 75-80c. I'm guessing I would need a new cooler to give it more vCore and VTT?

[famish99]

Anyways, CPU-Z is reporting a vCore of 1.280 and my VTT is set at 1.375 in the BIOS. Should this stress test fail, should I give it more vCore or more VTT? My temps are hovering around 75-80c. I'm guessing I would need a new cooler to give it more vCore and VTT?

Well you're supposed to have found out whether you needed more VTT first by doing the bclk experiment. If your system ran stable at a low core multiplier and high bclk, then experiencing any failures at high multiplier should most likely be a function of Vcore, there are always exceptions, occasionally at the higher frequencies, if the frequencies are close enough or harmonic of, you can get coupling or crosstalk issues which may cause you to raise both Vcore and VTT, but usually not.

75-80c is on the warm side, so if it's unstable, it's best to back off or get a new cooler because even though the die is rated to 100c, you can get spikes that are too quick for the thermal monitoring program to read that will shoot it past 100c and shutdown your comp.

[Chewie_77]

Check these settings out:

191 x 21 = 4011MHz
unCore Multi: x16
Mem Multi: x8
QPI Multi: x18/x36
vCore: 1.25000 BIOS / 1.264~1.232 Windows
VTT: 1.315 BIOS
IOH: 1.100
DRAM: 1.500



Again, I've been lowering my vCore and VTT since my last run last night.

[famish99]

That's some pretty low voltages, especially for the core, so I would want to do some more testing before continuing. LinX is somewhat bad at catching errors outside of the DDR and CPU so you might want to run some real world tests (or something like 3dmark) just to make sure that your IOH and QPI system are stable.

A fairly good test I've started to use is the World Community Grid because it's more real world application and unlike folding, provides very good fail data on their website. In addition, even if your system completes a work unit, but computes incorrectly, someone else will validate your work so you can find out if you're getting any errors that way as well.

[Chewie_77]

I've seen a few 4.0GHz 920's running with similarly low voltages, do you think I could actually go lower and still achieve stability?

I'm downloading 3DMark Vantage, now to find that key I have...

[famish99]

I'd test before you decide to go lower, I mean obviously good chips exist and it's always possible that you can go lower... but it's better to make sure you're completely stable where you are before you start going lower and finding yourself having to test your way back up.

[Chewie_77]

Ok, ran it no issues. Should I loop is a few times? Also, the only program showing my CPU speed is CPU-Z and Real Temp at 4011MHz. Other programs including Windows are off on the clock speed.



Edit: Just looped Vantage 3 times no problems, also, my IOH is 1.300, not 1.100 like I previously said, this fine?

Edit 2: Dropped the volts a bit on the vCore and VTT as well as IOH (now at 1.200), running some tests.