Core i7/X58 Overclocking Thread

[Hornet331]

So If I want to go over the 4Ghz "limit" I need to increase my DRAM voltage above 1.65 ?
And if I want to go above 1.65v I need to keep my QPI/VTT voltage at least 0.5v lower than Vdimm ?
Its getting a bit confusing i might say, what else is that it has no logical explanation to all that

BTW, I had no problems running my QPI/Dram Core (Asus Terminology) at Voltages up to 1.65v on P6T Deluxe.
Just haven't tried higher than that as it seems to have no effect above 1.6v and above 1.65v is too much voltage IMO.
I am watercooling my chip pretty good, so far My max temps are about 60-65c on the hottest core, so extreme high temps are not an issues for me.
My full spec is at my sig below.

You you got that right, if you increase the vdimm you have to watch that your qpi vlotage is not lower then 0.5V then your vdimm voltage.

There are people that run Qpi voltage 1,4V+ so i dont think its gona be a big problem. But you dont really need uncore clock that high, cause the gained performance is only benchable but thats all. You wont notice any difference while working.

[Kensek]

You you got that right, if you increase the vdimm you have to watch that your qpi vlotage is not lower then 0.5V then your vdimm voltage.

There are people that run Qpi voltage 1,4V+ so i dont think its gona be a big problem. But you dont really need uncore clock that high, cause the gained performance is only benchable but thats all. You wont notice any difference while working.

All the new Low Voltage Ram that runs at 2000 requires QPI/Dram vtt of 1.55-1.65v.

Intel specs say 1.35 Max on QPI Vtt
Dram Mfgs say the new Elpidia chips need up to 1.65v of QPI Vtt and 1.65v of DIMM.

Who knows?

[The-Fox]

You you got that right, if you increase the vdimm you have to watch that your qpi vlotage is not lower then 0.5V then your vdimm voltage.

There are people that run Qpi voltage 1,4V+ so i dont think its gona be a big problem. But you dont really need uncore clock that high, cause the gained performance is only benchable but thats all. You wont notice any difference while working.

I know it won't be noticeable, its just for Benchmarks, not 24/7 work.

OK, I went to try what you said above and now I am having a very weird issue.
I am trying to work with DRAM multipliers higher than 8 but I fail to post

For instance, I set bclk to 180, DRAM to 1800Mhz, and Uclk to 3600 (twice as dram).
I can't post, no matter what DRAM voltage I use, I climbed from 1.8v all the way up to 2.1v and I couldnt post.
It seems that either using uclk multi higher than 16 or DRAM multi higher than 8 is impossible for unknown reason.
I actually tried that on the newest BIOS came out about a week ago, 1102 But I don't know if its due to that.

Am I missing something here or is that a known issue/limitation ?

I tried to raise the QPI link to 22/24X, it seemed like it was really doing it but when I opened CPU-Z i found it was actually still 18x, is that normal ?

Help/ideas will be much appreciated.

[dejanh]

I know it won't be noticeable, its just for Benchmarks, not 24/7 work.

OK, I went to try what you said above and now I am having a very weird issue.
I am trying to work with DRAM multipliers higher than 8 but I fail to post

For instance, I set bclk to 180, DRAM to 1800Mhz, and Uclk to 3600 (twice as dram).
I can't post, no matter what DRAM voltage I use, I climbed from 1.8v all the way up to 2.1v and I couldnt post.
It seems that either using uclk multi higher than 16 or DRAM multi higher than 8 is impossible for unknown reason.
I actually tried that on the newest BIOS came out about a week ago, 1102 But I don't know if its due to that.

Am I missing something here or is that a known issue/limitation ?

I tried to raise the QPI link to 22/24X, it seemed like it was really doing it but when I opened CPU-Z i found it was actually still 18x, is that normal ?

Help/ideas will be much appreciated.

I have the R2E and I found running Uncore below QPI does not work too well on my board. Try running QPI with 22x multiplier when using 180MHz BCLK. 24x multiplier will not work as it is too high (24x180MHz > 4GHz which is max for QPI).

[Splave]

its a shame my chip hates HT so much

It does like playing dualie though

http://valid.canardpc.com/show_oc.php?id=477470

[tjelaw]

You you got that right, if you increase the vdimm you have to watch that your qpi vlotage is not lower then 0.5V then your vdimm voltage.

There are people that run Qpi voltage 1,4V+ so i dont think its gona be a big problem. But you dont really need uncore clock that high, cause the gained performance is only benchable but thats all. You wont notice any difference while working.

I run a vDIMM of 1.65

I'm trying to find the QPI / Uncore / CPU VTT (they are all the same right?) voltage, but I cannot find it. I'm using an Asus P6T Deluxe, anyone know what this setting called on this board?

I do have QPI/DRAM @ 1.4... but dont think thats the same setting or is it?

[The-Fox]

I have the R2E and I found running Uncore below QPI does not work too well on my board. Try running QPI with 22x multiplier when using 180MHz BCLK. 24x multiplier will not work as it is too high (24x180MHz > 4GHz which is max for QPI).

The Uncore is running faster than QPI, maybe thats what you meant.
I tried using Multi x22 as well as X24 but it will actually give me x18 regardless of the multi I choose, at least that is what GPU-Z is telling me.

I saw a lot of people out there who run high Uncore speeds, their Uncore is way faster than the QPI so it is possible.

Can anyone try that with ASUS mobo and see if that works for him ?

[Tobit]

Received my 920 today, it's from batch 3839A579 - unfortunately, I am still waiting for my motherboard. NewEgg is still having some issues.

[Hot'n'Chilly]

Received my 920 today, it's from batch 3839A579 - unfortunately, I am still waiting for my motherboard. NewEgg is still having some issues.

If NewEgg is slow you should complain to them via email. If you have a legit gripe they will give you cash off your next order.

[Hot'n'Chilly]

Hey guys, I've noticed a lot of low-voltage people having SpeedStep turned off. And idea what SpeedStep does, or if this might help out?

[scaramonga]

Hey guys, I've noticed a lot of low-voltage people having SpeedStep turned off. And idea what SpeedStep does, or if this might help out?

SpeedStep

[Xello]

http://en.wikipedia.org/wiki/SpeedStep

It lets the cpu clock down (mhz) and power down (volts) when idle, to save power, if set properly in windows.

[bowman]

And, at least on the P6T, you can't have Turbo Mode and not SpeedStep. Disabling SpeedStep disables Turbo as well.

[mremulator]

Guys, I'm going to order an i7 940 tomorrow and swap out my i7 920. I'm convinced that the BCLK set to anything above 191 on my Asus RE2 is causing me problems.

I think that going from 20x190 plus HT enabled to 21x190 HT enabled, will enable me to go from 3.8GHz/4.0GHz to 4.0GHz 4.2GHz prime stable (where I really want to be). This obviously can't be done with the i7 920.

Lowering my memory multi and/or raising my Vcore seems to make no difference when @ 20x200. However, lower the BCLK from 200 to 191 and my rig will prime (small FFTs) for 14.5 hours+ (have not tried longer). Obviously, my DRAM, QPI and Uncore frequencies are slightly lower (1523, 3438 & 3056), however, the 8:9 ratio still remains. 1.4v Vcore, 1.35v QPI, 1.8952v CPU PLL & 1.65v VDimm by the way. Everything else set to AUTO.

dejanh is absolutely spot-on with his 20x200 observations, as I have had little success with this combo. I really am starting to think it's more about frequencies/multipliers than voltage now. I think I have found the best voltages and timings for my board & RAM, just need to get a 'slightly' higher multiplier me thinks now?

Will find out in a couple of days.

[avrionis]

Which temp should I be looking at ? The main CPU temp or the cores when using CPU-Z?

Which temp are others referring to, when they have posted on this thread?

Also, my Main CPU temp is about 3-15 deg's more when comparing to 1 of the 8 core temps?

[SteveRo]

Which temp should I be looking at ? The main CPU temp or the cores when using CPU-Z?

Which temp are others referring to, when they have posted on this thread?

Also, my Main CPU temp is about 3-15 deg's more when comparing to 1 of the 8 core temps?

Folks use different temp read out tools.
Most folks, including myself use latest version of real temp - an excellent tool that I trust.

[speckled10]

Which temp should I be looking at ? The main CPU temp or the cores when using CPU-Z?

Which temp are others referring to, when they have posted on this thread?

Also, my Main CPU temp is about 3-15 deg's more when comparing to 1 of the 8 core temps?

A $64 million dollar question indeed !!

Unclewebb sums it up pretty well in Post #3021 here ..

http://www.xtremesystems.org/forums/...79044&page=121

Basically the cores temps are the ones to trust - they are designed & calibrated by Intel to safeguard your CPU from too much heat - but the sensors are only accurate at high temps .. and even then nobody has a clue what the absolute true temperature is (since actual Tjmax is unknown for each core) ..

If you have an i7 though - the sensors have improved in accuracy right down to Idle temps .. and if you follow Unclewebbs calibration guidlines for Realtemp you can get pretty good readings ..

As for CPU Case temp .. If your readings are that much higher sounds like you have a similar problem to me .. ASUS PC Probe totally screws my CPU temp readings .. so I dunno which program to trust ..

[rge]

Which temp should I be looking at ? The main CPU temp or the cores when using CPU-Z?

Which temp are others referring to, when they have posted on this thread?

Also, my Main CPU temp is about 3-15 deg's more when comparing to 1 of the 8 core temps?

Core temps are calibrated at the factory by intel and are read directly from the cpu by programs like real temp, and are the only temps that matter (especially on core i7's where they are more accuate), as they are the temps that intel calibrated to protect the cpu.

CPU temps are calibrated by the bios (from intel specs) as a best guess/compromise of IHS temp, but it will neither be accurate of core nor accurate of IHS temp, because the sensor is located between the cores and the gradient to the IHS or core varies based on load. Add in all the other errors, electrical interference, etc and then you know why intel abandoned cpu temps in favor of core temps for prochot. A future bios update may correct your obviously incorrect cpu temp (cant be higher than core temps) to a just as inaccurate but more plausible one (makes the end user happy), but gives just as useless information on core i7. I personally would just ignore the cpu temp and go by core temps, like intel does.

[eva2000]

1. VDIMM has to be within 0.5v of Uncore voltage (refered to on some boards as QPI/PLL volts or CPU VTT) for CPU to be safe above >1.65v DRAM memory voltages True

2. CPU VTT max 1.375v spec'd some have benched 1.55v so gives VDIMM a 1.875-2.05v range FALSE
3. UNCORE (CPUZ NB frequency) needs to optimally be twice that of MEM frequency TRUE
4. SOMETIMES for stability, having VCORE within 0.02-0.03v of CPU VTT might stabilise things Way FALSE
5. Yes, theoretical QPI frequency limit is around 4000Mhz but it's too early to say if we can surpass that, some boards and cpu combinations possible could in future. Way TRUE

i'm going by my own first hand experiences regarding #4. So what's the right answer for #2 ?

[Hornet331]

Core temps are calibrated at the factory by intel and are read directly from the cpu by programs like real temp, and are the only temps that matter (especially on core i7's where they are more accuate), as they are the temps that intel calibrated to protect the cpu.

CPU temps are calibrated by the bios (from intel specs) as a best guess/compromise of IHS temp, but it will neither be accurate of core nor accurate of IHS temp, because the sensor is located between the cores and the gradient to the IHS or core varies based on load. Add in all the other errors, electrical interference, etc and then you know why intel abandoned cpu temps in favor of core temps for prochot. A future bios update may correct your obviously incorrect cpu temp (cant be higher than core temps) to a just as inaccurate but more plausible one (makes the end user happy), but gives just as useless information on core i7. I personally would just ignore the cpu temp and go by core temps, like intel does.

Dont forget that the core temp shows always the hottest part on the core. Ci7 has several DTS on each core, but the PCU only outputs the worse Core temp. That also might explain, why we see relative high idle temps compared to C2D.

[Kensek]

i'm going by my own first hand experiences regarding #4. So what's the right answer for #2 ?

Answer to No 2 is Not Sure.

Though Intel says the Max QPI Vtt should not exceed 1.35v all of the new Elpidia DDR3-2000 to DDR3-2333 Ram that need to operate at a QPI of 1.55- 1.65+ volts.

The G. Skill black pi I'm using now operates at DDR3-2021 at 8-8-7-20 1N with 1.585 QPI and 1.70 VDimm.

The Corsair DDR3-2000 7-8-7-20 GT needs 1.65v on QPI and 1.65 v Dimm

What will the long term effects be?

All the new fast DDR3-2000 Ram that run at 1.65vDimm need 1.65 QPI volts also.

This is needed since Uncore must run at 2X DDR3 Frequency.

[ReverendMaynard]

its a shame my chip hates HT so much

It does like playing dualie though

http://valid.canardpc.com/show_oc.php?id=477470

Eff HT, you're hitting some nutty clocks bro.

[rge]

Dont forget that the core temp shows always the hottest part on the core. Ci7 has several DTS on each core, but the PCU only outputs the worse Core temp. That also might explain, why we see relative high idle temps compared to C2D.

Since Digital Thermal Sensors were first used by intel, intel has widely documented that DTS logic scans several transistors functioning as diodes on die (located in hot spots, ie L1, DTLB, IntReg, etc) and stores the maximum temp in MSR which is then accessible by realtemp, etc. It has always been that way, only difference on core i7 is interface, etc, not use of multiple sensors.

If you drill a hole through IHS, and through surface of die and embed a thermocouple, you can get reasonable accurate idle temps on core 2 duos, that are always 7-8C above ambient, even at low volts and idle. But if core sensors (located in hotter part of die than a thermocouple on top of die) read lower than ambient, it tells you those sensors were not placed poorly, they just were reading very inaccurately at low range. This has been confirmed by intel engineers multiple times, though that was already clear, that they are not linear at low range. The same intel engineers that clearly stated the core i7 would have more accurate and more linear sensors, it is simply a matter of using more expensive ones. You can go to sites like national semiconductor and read about pricing versus linear accuracy on DTS, all a matter of beancounters willing to pay for the slightly more expensive ones that will maintain linearity over a wider range, to appease the public, as intels purpose is only served by making them accurate at high end for throttling.

[Hot'n'Chilly]

Strange problem I just noticed on my P6T bios with my 920. It won't let me raise the CPU Ratio (miltiplier) above 20. It says something about a min od 12 and a max of 20. Is there a way of getting around this???

[DavidNJ]

Well, , a 940 would let you raise it to 22? Or a 965 to 256 I believe. Turbo effectively gives you 21. And turning off hyperthreading I believe gives you 22. However, it may not give you better performance.