Pontius,

ok, so does the Intel Burn In Self Test crash, hang or blue screen your system, or just give you results that indicate instability?

actually, nevermind that question...

So, since you are making it into windows and can run the intel burn test to some degree, you are hitting mild to moderate instability. In order to tune GTLs, you will need to keep pushing the host clock higher @6x without raising the vCore any (increasing vFSB and vMCH are fine).

Like I said before, there is a progression of overclocking to take here that results in having all areas stable, but you have to push the components in a particular order so that "interfering instability" from other components is minimized...basically making your life easier. and again, they are:

1. with the cpu at its lowest multi and your ram slow and loose, either
a) max out the vFSB and the vMCH (you will be fine here since your CPU and MCH are on water) and start increasing the host clock until the system freezes on boot. Write down the host clock at which the system freezes on boot. Since we are running a low multi, we are (in theory) finding the "maximum" that your FSB can run at.
or b) increase the host clock with vMCH and vFSB at default until instability occurs, then raise vFSB a notch or two...if the instability is gone, continue increasing the host clock until unstable again. if instability remains, increase the vMCH a notch or two and see if it helps. Keep doing this until you reach voltages that you do not want to go above. similar to above, this is the "maximum" that your FSB can run with the selected voltages.

2. once you know the "maximum" FSB for your selected voltages, you can begin tuning the GTLs found under "Reference voltage Override" in the CPU page of the BIOS. there are 3 GTLs that you will primarily be concerned with, 2 here (CPU GTL REF0 and 1), one later (MCH GTL REF), and possibly MCH DDR REF (not much performace or stability to be gained here, though.)

make sure you write down the values listed when you change from auto to manual. you will want to change them one at a time and they will most likely need to be raised, but this depends on the internals of your FSB and CPU cores, so you will need to go both up and down to find the best setting.

One thing to remember here is that while this fine tuning allows for higher FSBs and CPU clocks at lower voltages, the results you will see while testing are rather unimpressive. So with the host clock at the setting where windows freezes on boot and without changing the multi or the voltages, increase the first GTL ref by one. save, restart, and again count how many times the progress bar goes by on boot. now go back to BIOS and change the same GTL ref down one from default. save restart watch the bar.

now that you have gone both up and down one from default, you should be able to tell which direction increases the amount of times the status bar goes by before crashing. if you cannot, go 2 up and 2 down from default and compare. Now that you know which direction to go, change the ref by 1 in that direction save and reboot, count status bars, increase again, count status bars....once you increase beyond the optimal, the number of times the bar passes will start to go back down. then you know you went too far, change the setting back to the level that gives you the longest boot before crashing.

Now move on to the CPU GTL REF1 and repeat the same process. if changing these settings allows you to enter all the way into windows, simply increase the host clock until it freezes on boot.

Once both CPU GTL REF0 and 1 have been tuned DO NOT CHANGE your vFSB.

3. Now having found the optimal cpu refs, increase vCore some and the system should boot @6x multi. Start increasing the host clock again followed by increasing the vCore until you reach the point where you do not want to increase voltage anymore. this will give you the true maximum FSB (well, almost true, unless your vCore and vFSB are maxed).

4. now you know pretty much exactly how fast your FSB can go and can start tweaking memory. with all voltages untouched from #3, drop your host clock back to 333 and, while keeping memory timings loose, change the memory reference frequency to 333 and the memory frequency to 1333 (this is pretty much the fastest practical combination that the x38 as configured in this board can handle) and save and restart. once into windows, run burn in test at least 6 times and make sure that all numbers are the same.

***note*** On the x38/48 boards from intel, resulting memory speeds as seen on the first screen of the BIOS page cannot be higher than the FSB speed.

***also note***this board is f-ing stupid when it comes to changing memory settings....often times you will need to restart several times or even cut all power and then restart a few times to verify that the system will not boot with a particular setting.

5. if burn test results are good, increase the host clock by 20mhz or so and repeat testing. if your memory are good quality, you can probably get the mem frequency into the 1800s (at 333x1333 you will reach this at 450mhz host clock), but you may need to do some research on your ram to see how fast others have gotten it.

as you are working your way up and you start to have issues with booting, or instability or crashing during burn test, you will need to increase the vMCH to correct it. if you have reached either the max vMCH or the maximum you want to run, then you can start playing with MCH GTL REF in the same way you changed the CPU refs. you should know, though, that there is much less functional adjustability with the MCH GTL than with the CPU refs...a few steps is all you will be able to change and see a difference, but those small changes will result in very large gains.

6. once you max out the memory speed at 333x1333, be sure to write down the frequency and the host clock numbers for your reference later. also note that it is unlikely that you will be able to run at this speed at higher multi's, but at least you the max that either your memory can run or the max that the MCH can handle.

Next you will want to go to BIOS, drop the host clock by 20 or so, and then change the memory ref and speed to 400x1600 (results in the same DDR3-Speed, but internal timings of the MCH are loosened). this may or may not boot, but if it does, start increasing the host clock by 5 or 10 and run burn test. repeat until you find the max host clock and memory frequency at this memory setting and write down.

7. continue to work your way down in memory speed....333x1333, 400x1600, 400x1333, 333x1066, 266x800 (unlikely to boot), 400x1066, and finally 333x800 (you already know the max for 400x800, so you dont need to do this again.)

8. now that you know the fastest FSB and the fastest that your memory can run with different memory reference frequencies and memory frequencies, it is time to test the potential of the CPU.

so, set your memory back to 400x800, and go to the max host clock that you found in step #3. save and reboot just to make sure that it still boots. once into windows, restart, go to BIOS and increase the multi by 1 step. save, restart, and run burn test. if everything is ok, restart, go to BIOS, and increase again. Keep doing this until it isnt stable or wont boot.

when you get to the instability part, either increase vCore if you are comfortable doing so, or reduce the host clock until stability is reached.

do this all the way up to your highest multi and write down the highest host clock for each multi and the resulting CPU frequency as you go.

you will find yourself reducing host clock frequency quite a bit as you get closer to your max multi. for example, my q9550 @ 6x could boot at around 505Mhz x6, but at 8x somewhere around 490 was the highest that would boot and the highest stable at 8.5x is 485 and that is with a chit ton of vCore...1.565 or so.

9. now that you know the maximum processor speed at each multi, look at your memory testing and see which setting would run at or above that host clock. pick the fastest resulting DDR3-speed, set it up in BIOS and drop your host clock by 20-30 but stay at highest multi. save, restart and see if it boots. if it does, run burn test 6-10 iterations and see if it is stable. if it is, then increase host clock by 5 or so and try again. if it isnt stable, decrease the host clock and try again. once you know the fastest memory speed that the MCH can handle at your highest multi you can decide if that is good enough or if you want the CPU to be faster and the memory a little slower.

if you choose the faster cpu, leave your host clock at the stable from #9 and change memory settings to next fastest from step #7. save, restart, burn test. if good, increase host clock a few and repeat.

Once you get to a combination that you like that is "stable" run 30 iterations of burn test and see if it passes. if it does, you can start reducing memory timings (start with going from Command Rate of 2T to 1T and see how it goes) if ok, lower the first timing option in BIOS then 2nd and 3rd...if command rate 1T just crashes system (and it probably will) decide if you want faster memory speed or tighter timings and either stay at current mem ref and mem freq and reduce timings with Command Rate at 2T, or reduce host clock until 1T is stable, or reduce memory ref and memory frequency until 1T is stable at the higher host clock.


ok, i am done for now. this should keep you busy for quite some time.

as far as voltages go, yes, increasing them will shorten the life of the component for which they are increased, but whether you will still have this board and processor by the time it dies is unlikely. the most common issue with operating at speeds requiring high voltages is not board or component failure, but rather a degredation of stability at those voltages, requiring you to either increase them more, or overclock to become stable again. but like i said before, my system is running at max vFSB, vMCH, and 1.565 vCore for 6+months. No degredation, no need to increase voltages further. I will for sure let all of you know when these issues come up.

on that note, the 1 voltage to be careful with is the vCore. I havent read anything about peoples processors being killed by high voltage so long as good water cooling is installed. That being said, i wouldnt go over 1.5 if you are at all concerned about killing your CPU...if you are less concerned and may buy another computer in the next year, stay under 1.6.