Core i7/X58 Overclocking Thread

[unclewebb]

Cpu-z though reports 4Ghz as if it were a constant frequency, whereas Real-Temp reports many variations, with the multi covering all the range from 20 to 25.

One thing I mentioned to rge is that how Windows is set up also plays a part in your actual multiplier at idle. If you have C1E and Speedstep enabled in the bios, you will also need to have your Power Options set appropriately so your multi isn't jumping up and down at idle.

In XP set your Power Options like this to get the full effect of C1E and the lower power states:
The picture is of my Q6600 dropping from 9.0 to 6.0.



For Vista you need to reduce your Minimum Processor State:



If RealTemp is reporting that your multi is jumping up and down it's because it is jumping up and down. I decided to let RealTemp start telling it like it is and then users will at least know exactly what's going on and they can decide what needs to be adjusted in the bios or Windows or both.

Edit: Another thing that rge has noticed is some of the numbers displayed by CPU-Z simply don't add up. It seems to be using multipliers of 1/2, 1/4, 1/8 or even 1/16 internally in some of its calculations:



196.0 X 21 does not equal 4111.5.
It equals 4116.0 MHz

[freshy]

even when i was priming @ 4.2ghz with turbo on , HT on my temps just hit 90, and that was with these voltages and i did not see any throttling , but i have cpu thermal monitor disabled in bios (bios f4k) on ud5

ok voltages for 4.2ghz 8 hours 38 mins of prime
200 x 19 + turbo, HT on
cpu vcore 1.55v
qpi/vtt 1.535v
cpu pll 1.88v
qpipll 1.18v
ioh core 1.24v
ich i/o 1.58v
ich core 1.22v
vdimm 1.66v
qpi link was x36
and uncore was running at 3400mhz

screen shot was taken while @ 4.1ghz with turbo on HT on but i have it set the same for when i was priming at 4.2ghz, anyway i think im gonna flash to f4l now and give that a go

[nutcase]

This is my i7-920 presently on a P6T deluxe. presently running 2 260 Gtx 192 core Cards in SLI.

Cpu is water cooled with swiftech GTZ WB

[Splave]

is there any performance benefit or detriment to running 6x1gb sticks? Like I know the octa-core mac pro benefits from having multiple 8x2gb rather than 4x4gb sticks

[dejanh]

Here is my i940. I forgot to take a picture of it after 8 hours of prime95. Highest temp was 88 during that run.
vcpu 1.325
pll 1.81
qpi 1.39

This is getting just stupid. Could it really be my cooling that is causing me to have to use such high volts for the same clocks? Just on this page there are 4 people running 4GHz+ overclocks with i7 920/940 with voltages between 1.32V and 1.36V. I cannot get true 4.0GHz stable in any combination of BCLK x multi unless I use 1.416V (CPU-Z). Even for 3.96GHz I need 1.40V.

Don't get me wrong, I am not complaining about the OC (I did expect more though out of the i7 940), but about the volts needed to get the OC. Has anybody gone from air to water and confirmed that volts can drop for same clocks? I have seen a voltage drop for same clocks going from stock air to aftermarket air, but is the same true for air to water?

[gooddog]

This is getting just stupid. Could it really be my cooling that is causing me to have to use such high volts for the same clocks? Just on this page there are 4 people running 4GHz+ overclocks with i7 920/940 with voltages between 1.32V and 1.36V. I cannot get true 4.0GHz stable in any combination of BCLK x multi unless I use 1.416V (CPU-Z). Even for 3.96GHz I need 1.40V.

Don't get me wrong, I am not complaining about the OC (I did expect more though out of the i7 940), but about the volts needed to get the OC. Has anybody gone from air to water and confirmed that volts can drop for same clocks? I have seen a voltage drop for same clocks going from stock air to aftermarket air, but is the same true for air to water?

I wonder how much is dependent on VQPI. Some of the better OCs I have seen seem to pump VQPI way beyond the 1.35 limit Intel recommends.

[nutcase]

I have found keeping QPI frequency under 3800 Mhz helps the overclock a lot.

[dejanh]

I wonder how much is dependent on VQPI. Some of the better OCs I have seen seem to pump VQPI way beyond the 1.35 limit Intel recommends.

Yeah, you do not want to do that really. Not too much anyway. That's probably the #1 way to kill this chip.

[JacobF]

Pictures speak louder than words

[Cooperdale]

With speedstep enabled, Realtemp will report varying multis depending on background activity of computer ie cores active vs not. CPUZ reports constant multi with speedstep enabled, ie does not seem to work correctly with speedstep enabled. If you disable C1E and EIST, then both Realtemp and CPUZ should report a stable multi.

Actually Speedstep is disabled, that's the strange thing. I will double check into the bios though. I've saved and loaded so many profiles, I may have missed something.

I wonder how much is dependent on VQPI. Some of the better OCs I have seen seem to pump VQPI way beyond the 1.35 limit Intel recommends.

Yeah, you do not want to do that really. Not too much anyway. That's probably the #1 way to kill this chip.

In my setup, OC is incredibly VQPI dipendent. I am running stable at 160x25 (20 hours of Prime95, 4 hours of Prime95+LinX, 10 runs with two instances of LinX running on half the memory each and with affinity to 4 cores each). I managed to lower temps on air to rare peaks of about 83°, with an average of about 80°. Not the most silent rig I ever had though .

Vcore is 1.34375 and VQPI is at 1.35, and I think I'm safe here, but if I lower it a bit, say 1.325, Windows will freeze at boot! Without touching any other setting.

If I raise the BCLK, I can't get stable even with higher vcore, unless I raise VQPI to almost 1.4V. I decided to let it go since my aim was 4Ghz and I'm there. But I hate how so many softwares report wrong cpu speed even at load. Can't fix them all I guess.

[Hornet331]

LOL...interesting, and you also have GB extreme? Crazy how everybody is seeing different things with different bioses, setups, OS, etc. Are you on vista? I just enabled speedstep and checked again, Realtemp fluctuates and reads lower multis, cpuz stays rock solid at bios setting 21 multi, but I am on XP...apparently something different between my setup and yours.

Yep im on vista 64 and using the Extreme.




in the second picture, igonre the 12x multi on realtemp. Realtemp shows the switches a lot faster then CPU-Z does.

Im on F4j and have enabled everything in advance cpu settings beside c5/6/7.

[rge]

^^It looks like speedstep with core i7 is working much better on vista 64, than on my xp sp3 32...also looks like cpuz is working better on vista 64. Multis on mine with speedstep (realtemp) vary down to around 15, but dont get lower. They only go to 12 (cpuz and realtemp) if enabled laptop settings, ie force the lower settings....then they stay on 12 unless loaded.

[Hornet331]

seems so, xp shows its age regarding to hardware.

I was a longtime xp user my self, but with 12gb i decided to switch to vista.

[unclewebb]

seems so, xp shows its age regarding to hardware.

When using Vista, do you have to go into the Power Options and reduce the Minimum power state so that RealTemp reports a 12X multi at idle? If this is set to 100%, what does RealTemp and CPU-Z show?

I think the two OS have just changed how they handle C1E and how they need to be set up to get the multi to drop to its lowest value.

[Hornet331]

When using Vista, do you have to go into the Power Options and reduce the Minimum power state so that RealTemp reports a 12X multi at idle? If this is set to 100%, what does RealTemp and CPU-Z show?

I think the two OS have just changed how they handle C1E and how they need to be set up to get the multi to drop to its lowest value.

if its set to minumum 100% it will show always full multi.

[rge]

My final 24/7. 4.2Ghz (200x21) with 1.412 bios LLC, 1.376 load, prime 12+hrs.
QPI/UC/Mem 36/17/8
QPI/Vtt 1.335
Mem 1600mhz 1.64v
CPU PLL, QPI PLL +1

[unclewebb]

One problem that software has with reporting the correct multiplier is that there are situations when during the standard 1 second reporting interval, the turbo might be engaged for three quarters of a second and disengaged for a quarter of a second. If you have a 940 and your base multiplier is 22 then your average multiplier is going to be 22.75 for this reporting period. How should this be reported is the big question?

This might be when CPU-Z starts reporting 22.5 for the multiplier. Technically that might not be correct since half multipliers don't exist but maybe users would like to see that since it tells them that their turbo is starting to become disengaged. When RealTemp sees an average multiplier of 22.75, it rounds that off to the nearest whole multiplier which is 23. In this situation, RealTemp is telling users that nothing has changed and the turbo is still fully engaged.

The question is, what do users want to see? A half multi that isn't technically correct or some other sort of warning that turbo mode has started to become disengaged?

I re-did my turbo testing tool and it's available here.
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

The FileDen was down for a second or two so I also uploaded it here:
http://www.filesavr.com/turbo

This tool calculates your exact average multiplier and displays the results to 3 digits. With a 940 and turbo engaged it should be pretty steady at 23.000

If the turbo starts to become disengaged due to heat or whatever, it should be able to detect that very quickly. I have no idea when CPU-Z will switch to 22.5. Time for some testing to get this mystery sorted out.

This tool is designed for Core i7 but it works with any Core processor. My Q6600 is pretty steady with this tool at 9.000 when fully loaded or it sits fairly steady at 6.000 when C1E / SpeedStep are enabled.

[Splave]

Going to try getting the ram back up to 1600 now
We'll see how that goes.

[vlocker27]

hi guys,

i just finished putting system together yesterday and everything looks fine with auto bios setting except video crashes sometimes when trying to view youtube full screen. anyways, thats not why im here. im here to see if i can get some help OC my system for image/video editing and some gaming. since i put out a lot of cash for this new system it's pointless to run them at default setting

im an OC newbie so i dont want to go extreme. any tip on on putting my 940 to 3.4-3.6 and ram to 1600 would be greatly appreciated

system specs:

- Antec TruePower Quattro TPQ-1000 1000W ATX12V
- Asus Rampage II Extreme x58
- Intel Core i7 940
- Thermalright "Limited Edition" TRUE Black Ultra-120 eXtreme CPU Heatsink
- 6gig of Corsair Dominator DDR3 1600
- Sapphire Radeon HD 4870 1gig GDDR5


thank you in advance

[Splave]

hey vlocker you might want to take a gander @ http://www.extremetech.com/article2/...129TX1K0000532
Atleast get a feel for what you are doing.

[dejanh]

Update from my end...I managed to get 23x179MHz stable after putting on a TRUE instead of the V8. Temps are about 5C-10C lower overall when chip is 100% loaded for extended periods of time. The voltage that is required for this is unfortunately high at 1.46250V BIOS (on R2E this is 1.49250V measured load roughly). After everything, I have concluded that I have a high-voltage chip unfortunately and this is limiting my max overclock. With a lower voltage chip I am sure I could break the 4.3GHz mark, possibly even on air. Temps right now at 4117MHz peak out at 83C on cores 1 and 2, and 81C on cores 3 and 4. This is with 4 cores and 8 threads enabled, all settings Auto except core voltage. Memory is running at 1794MHz with 8-8-8-24-2T timings for my 1600MHz Dominators with 1.67007V in BIOS (1.652V measured load). Basically, I had to back off 1MHz off BCLK to be stable as 23x180MHz would not be stable under any circumstances (I think I fall victim to the runaway amps due to the high voltage and heat when I am running any higher than 23x179MHz).

Overall, alright on air

Going to be putting this chip on sale most likely with the V8, and a couple other goodies with it. I will be offering a reasonable bundle price so if anyone is interested PM me and we can talk

[bmilos]

what are you going to pick up to replace it? I was thinking about possibly doing the same...

Also, I am having memory issues. Anyone else have A 6gb kit of XMS3 1600? I have a hard time running it even at 1600.

[Hornet331]

Ci7 is the first chip that makes me think about getting a water cooling setup.

Cause i think 4.2ghz is possible with my chip with reasonable temps, but aircooling dont cuts it.

[bmilos]

This is also my first water cooling setup too. Its definitely the way to go, but it can get super expensive... The justification is that once you get it all together, the future just holds add-ons and upgrades..

I got the NBMAX from swiftech, but am thinking about picking up the EK full block.

Also, anyone using EK's MOSFET blocks? What are the gains using those in place of stock?

[dejanh]

what are you going to pick up to replace it? I was thinking about possibly doing the same...

Also, I am having memory issues. Anyone else have A 6gb kit of XMS3 1600? I have a hard time running it even at 1600.

I'm running the 6GB kit of XMS3 1600. I am Prime stable with them at 4.117GHz CPU, 1794MHz DRAM, and gaming stable at 4.140GHz CPU, 1804MHz DRAM.

Timings are 8-8-8-24-2T for now, though I will likely tighten them a bit. I always use the X.M.P. profile in Ai tweaker in the BIOS and then set the DRAM voltage on the R2E to 1.67007V in BIOS (actual measured is approx. 1.65V) and QPI/DRAM to Auto. If you are running 1600MHz+ DRAM your QPI/DRAM needs to be a minimum of 1.35000V in BIOS...at least for the XMS3 kit.