OC Walkthrough:i7 920+Prolimitech Megahalems, Intel DX58SO

You should read this guide. It helped me a lot. I'm only started with Intel 2 weeks ago and the guide helped me to understand principles of i7 overclock.

Also here's very nice calculator for all frequencies.
Edited by McDown - 9/18/09 at 6:37pm

Just noticed your friend request and it led me to this thread.

It's easy to get to 4.0 with this board. Don't listen to anyone that tells you otherwise. In the past Intel boards have not been good overclockers, but this board is a complete 180 degree shift away from that. In fact, this board can do some things with overclocking that no other board can match, like using the 22x multiplier for single cores while overclocking.

The biggest factor is the CPU. Is it a C0 or D0 stepping, and what batch # if you know? The # is on the side of the box and engraved on the last line on the chip.

I've been meaning to write something up about how to overclock the i7 since I learned so much the last month. I would start with finding the highest base clock your CPU and board are capable of. First set up these base settings so nothing could be causing instability except the setting you are testing:

CPU - 1.20 V, 13x, Turbo off
Memory - 6x, 1.66 V, 9-9-9-28 (or whatever the nominal settings are for your RAM)
Uncore - 13x, 1.35 V
IOH - 1.1

Start at 200 and go up by 5 up to 210, then by 2 until you find the base clock. You may have to start lower if you can't do 200. I got to 218.

Now that you know the base clock limit, you can figure out your max CPU speed. Change the multiplier to 20x and turn on Turbo and Current Override in the CPU section, and turn off C state and C1E in the last tab in the BIOS. This will get you a 21x multiplier which you want to be at when overclocking.

This board lets you control Vdroop via the power slope option. I set mine to 50% because I found that the default 100% resulted in a significant Vdroop under load. You can see this happen when the voltage drops in CPU-Z during start stress testing. The 50% setting made my idle and load voltage exactly the same or within 1 notch of each other. Keeping it steady is good to avoid surprises with voltage related instability but the power slope setting to get that effect probably varies with the CPU and board, so experiment.

Go to a base clock of 155 and do some stress testing. I recommend LinX 10 runs 10,000 problem size at this point just for the sake of speeding it along. You can do more extensive stress testing later. You should still be at 1.2 V so see how high you can go by increasing the base clock 5 notches. I think I got to 180 (3.8 GHz) on stock volts, but my CPU is cherry picked from a known good batch.

When you find your max at stock volts, increase the voltage by 0.05 V. So from 1.2 V go to 1.25 V and see how far you can go. In my experience each 0.05 V increase yields a 100 MHz gain. That translates to a 5 base clock increase. Once you get to 195 start increasing the volts by 0.025 V and the base clock by 2. You also want to start watching your temps more closely while stress testing because you are likely to hit 100 C over 200 base clock / 4.2 GHz if your heatsink is not mounted well.

The max speed I bothered to test was 210 (4.4 GHz) but it got too hot during stress testing so I scaled back to 206 (4.33 GHz) and got it stable at a resulting (in CPU-Z) voltage of 1.412 under load. I don't remember what the exact voltage is set to in the BIOS (it's probably a bit higher), but you want to be going by the CPU-Z voltage under load when logging your attempts.

After you've settled on a good speed with good temps you can start finding the QPI, which is the second most determining factor for performance after the CPU. Unlike when you found the highest base clock, you want to keep your CPU multiplier on 21x because QPI in my experience will go much higher on a lower CPU multiplier and this gives you false data. (I got to 4000+ MHz QPI on a 13x CPU multiplier with 1.15 V QPI voltage, but couldn't break 3500 MHz without increasing QPI voltage on a 21x CPU multiplier. I posted about this anomaly but no one commented, likely in part because few people pay attention to the QPI and only care about their CPU speed.)

Keep the base clock at the same point where you are happy with the resulting CPU speed and increase the QPI multiplier starting at 16x by 1 notch until you start to see instability during stress testing again. Raise the voltage by .05 V and try the same multiplier again. If it's still not stable raise by another .05 V until it gets stable. When it does, you can raise the multiplier by another notch and try to get that stable. I got to 18x at 1.325 V at 206 base clock and 19x at 200 base clock which is my 4.2 GHz 24/7 OC (voltage is 1.35 V but I haven't had time to try to lower it yet.)

Last is the RAM. It comes last because it also gives false data if your CPU multiplier is set lower than you will really be running at, and you need the headroom from a high QPI multiplier to raise the memory multiplier (if you haven't noticed the BIOS forces the QPI multiplier to be at least double the memory multiplier; this is because anything else causes instability due to the way the Nehalem architecture is designed.)

The RAM is highly dependent on your specific sticks. I sent back my first 1600 MHz kit because it couldn't perform at 1600 MHz when my CPU multiplier was 21x. I replaced it with a 1866 MHz kit and I am running it on my 200 base clock (4.2 GHz) 24/7 OC with an 8x multiplier. The memory multiplier only goes up in increments of 2 so I can't go to 10x because A) that would result ni 2000 MHz and my memory couldn't handle it, and B) my stable QPI multiplier is 19x, which would be raised to an unstable 20x as soon as I set the memory to 10x in the BIOS. Because I underclocked the memory to 1600 MHz I made up the performance difference by overclocking it to tighter 7-7-7-15 timings from the stock 9-9-9-28. In fact I probably surpassed the performance loss because of the aggressive overclock.

You want to use the same logic to find the max memory speed (and eventually timings, when you have time to mess with those) as QPI. If you need more clarification let me know and I'll elaborate, I'm just tired of typing at this point.

All of these factors will affect your OC. My modest 24/7 OC of 200 base clock, 21x CPU, 19x QPI and 8x memory results in about 56 GFlops in LinX with temps in the mid 80's.

Yes, you need to throttle those down when you start OCing. I did the same thing at the beginning and couldn't POST. The BIOS definitely has some quirks, but once you get used to it the hardware is top notch.

B-batch D0 will get you to 4.0+ easy.

Stack them. If you set them back down after you find the max, you will get falsely elevated maxes for the other settings. This will mess you up when you try to combine them, so you want to stack in that order: BCLK, CPU, Uncore and RAM.

See guide in my profile.