Understanding the X58 chipset Asus P6T BIOS Settings for the i7 Nehalem CPU

[c627627]

What's this BCLK thingy?

BCLK thingy times multiplier equals MegaHurts speed.


Kindly read & correct everything you see that doesn't look right.
The i7 runs hot. When running the Prime95 Torture Test for the old Athlon XP, alarm bells went off when the temperatures zoomed past 60° Celcius but I understand realistic OC threshold is now raised toward 90° for max stress when it comes to the i7 CPU.

Per Arctic Silver thermal paste instructions, we're supposed to have the i7 CPU golden triangle in the upper left corner, then apply one horizontal line of paste across the middle, apply NO paste to the heatsink and just mount like that.

Prime95 number of Torture Test threads should be 8 for the i7 CPU, not 4 like with the older CPUs. http://www.mersenne.org/freesoft.htm


I am using
Intel i7 920 CPU
3 x 1GB G.SKIL DDR3 1333 (PC3 10666) RAM
Thermalright Ultra-120 eXtreme 1366 RT heatsink with a 120mm Scythe S-Flex F fan
ASUS P6T Deluxe motherboard
Corsair CMPSU-750TX 750W


I'm looking to understand the following:

• What is CPU PLL Voltage vs. CPU Voltage?

• Should QPI/DRAM Core Voltage be raised alongside the CPU Voltage, pretty much identically?

• IOH vs ICH voltage vs IOH PCIE vs ICH PCIE Voltage? I plan to set the IOH to 1.2 Volts but what about the other three?

• DRAM Bus Voltage I set to 1.64, what is the effect of going over 1.65 the labels are warning us against doing.

• Load-Line Calibration should be [Enabled] but what exactly does it do?

• CPU Differential Amplitude, 700mV is the lowest, what exactly does it do?

• What is CPU Clock Skew vs IOH Clock Skew Delay. 100ps is the lowest, I understand it helps with BCLK overclock, exactly what we all need, but what is this setting about?


I'm writing this *before* I actually do the research on it myself so hopefully someone can spell things out and help along, and I'll pass your knowledge on .

Here are my preparations so far for the overclock:

- Think of CPU Ratio x BCLK Frequency as kind of like multiplier x FSB for the old Athlon XP.
- Do not overclock from within Windows as people report higher overclocks if overclocking from BIOS.

EDIT 6 [I'll be editing this as I figure out the answers (for my system)]


* MAIN

Storage Configuration > Configure SATA as: [IDE] ensures wider compatibility but [AHCI] allows the onboard storage driver to enable advanced Serial ATA features that increases storage performance
Set AHCI before installing the operating system, otherwise you need to make changes to your operating system if you want to switch to AHCI after installing the OS, more info:
http://www.overclockers.com/forums/showthread.php?p=6541941


* AI TWEAKER
Ai Overclock Tuner [Manual]

Intel (R) Turbo Mode Tech or in later BIOS versions: Intel(R) SpeedStep(TM) Tech [Disabled] This increases the multiplier by one if the temps are low enough but I figure the OC may not be as stable when it does, so I changed it to [Disabled]

CPU Ratio Setting [19] Locked above 20 but unlocked below. Reduce this setting initially to maybe [14] to find out your max stable BLCK Frequency.

BCLK Frequency [211]

DRAM Frequency [DDR3-1691MHz]

UCLK Frequency [3384MHz] Twice the DRAM Frequency.

QPI Link Data Rate [7615MT/s] The lowest setting. What is the definition of the QPI Link Data Rate?


DRAM Timing Control In my case I had to adjust memory timings increased from 8-8-8-20 to 10-10-10-24 to accommodate DRAM Frequency [DDR3-1691MHz]



CPU Voltage [1.40000] People tend to keep this at 1.35-1.4 since there's a disproportional increase in heat and power consumption when you go higher but in my case I may have a cool running/low current chip, and if so, it'll need more voltage than that.

CPU PLL Voltage [Auto] Helps with BCLK overclock but I'm not yet clear on why/if I should go up.

QPI/DRAM Core Voltage [1.35000] Analogous to northbridge voltage. Only needs to be increased for high memory speeds. In my case 1.35 volts is plenty for DDR3 1333. However even though this is Uncore or QPI/Memory controller voltage not the actual memory voltage, this voltage needs to be increased to within 0.5 Volts of the DRAM Bus Voltage to keep the CPU overclock safe and stable.

IOH Voltage [Auto]
IOH PCIE Voltage [Auto]
ICH Voltage [Auto]
ICH PCIE Voltage [Auto]

I understand now that the PCIE bus goes from ICH to IOH and "those would just be the voltages for the two links "so I'm leaving the IOH/ICH PCIE voltage on Auto. Still not clear on why/if ICH voltage should be left on Auto.

DRAM Bus Voltage [1.64] ... As long the difference between DRAM Bus Voltage and QPI/DRAM Uncore Voltage is less than 0.5 volts, we can ignore Intel's warning and go over 1.65 Volts if needed.

Load-Line Calibration [Enabled] Since it helps to "counteract Vdroop on the CPU."

CPU Differential Amplitude [Auto] Still not sure about this setting.

CPU Clock Skew [Auto] It appears it should be left on Auto, but need more info on this setting.

CPU Spread Spectrum [Disabled] To help with BCLK overclocks.

IOH Clock Skew [Auto] It appears it should be left on Auto, but need more info on this setting.

PCIE Spread Spectrum [Disabled] To help with BCLK overclocks.


* ADVANCED

Onboard Devices Configuration > High Definition Audio [Disabled] (if using Sound Card instead of onboard sound)
Marvell Storage Controller [Disabled] (For a faster boot but only if not using IDE and eSATA.)
Marvell SAS [Disabled]
CPU Speedstep [Disabled] (This option may not be there)

* POWER

APM Configuration > Power ON by PS/2 Keyboard > Space Bar


* BOOT
First, scroll down to: Hard Disk Drives > Only enable SSD, disable others
Second, go back up to: Boot Device Priority > SDD, DVD, 1st Floppy Drive, ATAPI CD-ROM
Boot Settings Configuration > Full Screen Logo [Disabled]

* TOOLS
Express Gate [Disabled]


* * * After you Exit & Save, the system will completely turn off - then pause a couple of seconds before starting up again. * * *



Note that it may not be overall the best idea to disable Hyperthreading so I left it Enabled.


CPU TM function (Thermal Monitoring) is the Auto-Shut down feature but what is the Auto shutdown temperature?


EDIT: If the i7 CPU golden triangle in the upper left corner, apply one horizontal line of paste across the middle, apply NO paste to the heatsink and just mount like that:

 

[Shiggity]

I know the load line calibration helps to counteract Vdroop on the CPU under load.

 

[Gautam]

QPI voltage is analogous to northbridge voltage. Only needs to be increased for high memory speeds. You won't need much more than 1.3-1.35 if that for 1600. 2000 is probably not practical if even possible. PLL voltage seems to help bclk overclocking out a little. The PCI express bus goes from ICH to IOH and those would just be the voltages for the two links. You probably don't need to touch any of these. (Unless you're overclocking PCI-E which is usually useless)

With clock skew settings the only way to find the best one is trial and error, but usually you'll be fine on auto.

 

[c627627]

↑ It's only the current world record holder Gautam posting.

I found out about more than half of things already, next up is:

QPI Link Data Rate [7615MT/s] The lowest setting. What is the definition of the QPI Link Data Rate?

CPU PLL Voltage [2.00] Helps with BCLK overclock but I'm not yet clear on why/if I should go to 2 Volts. Edit: No need, I tested to pass at Auto.

QPI/DRAM Core Voltage [1.35000] Although this is the Uncore or QPI/Memory controller voltage and not the actual memory voltage, this voltage needs to be increased to keep the CPU overclock safe and stable. So it does appear that there needs to be a correlation between this and the CPU voltage... They say this voltage needs to be increased to within 0.5 Volts of the DRAM Bus Voltage.

IOH Voltage [1.26]
ICH Voltage [Auto] Still not clear on why/if ICH voltage should be left on Auto and what exactly IOH Voltage increase is about. Edit: No need, I tested to pass at Auto.

DRAM Bus Voltage [1.64] Since I'm not going over [DDR3-1603MHz] for my DDR3 1333, why should I even change this from Auto?


CPU TM function (Thermal Monitoring) is the Auto-Shut down feature but what is the Auto shutdown temperature?


EDIT: I just started the 8 thread Torture Test and have case temp jump to 36° C and a CPU temp fluctuating above 87° C. I expect to hit 90° Celcius or so at the point of maximum stress. My idle CPU temperature was around what the case temperature was... These temperatures wold have killed my old Athlon XP...

 

[c627627]

200 x 20 @1.40 volts after 11.5 hours I started using it while Prime95 was running and it rebooted.

200 x 20 @1.45 volts after a few hours I started using it while Prime95 was running and it rebooted.


so now I'm not gonna touch it while Prime95 is running but I switched to 211 x 19 = 4009 MHz @ 1.50 volts.


I booted with BCLK at 220 and ran Memtest for a bit without errors, I guess I should have found out my highest BCLK first but what exactly is the advantage of having higher BCLK at same MHz speeds?

Is Higher BCLK worth a loss of a few MHZ and if so where's the break even point?

 

[ozzlo]

Those temps are for the individual cores, I have notices a wide range different temps acrossed all cores on my chip. I just average the 4 values instead of going off of the highest. My chip is currently Folding with HTT enabled and getting around 82c on the cores with a TRUE.

From what I have read > 1.65 mem volts can kill the CPU. I'm not 100% sure this is true but I have no reason to OC my mem that high anyways. With all the bandwidth from triple channel I have focused on tightening my timings instead of trying to OC the ram more.


For the life of me I can't get my board stable beyond 190 BCLK. I have tried several ranges of voltages and I have set all multipliers to their minimum settings so only the QPI is stressed and Prime 95 will kill it in about 10 seconds.

Even with the BCLK limitation on my board, I have managed to take advantage of the turbo setting boosting my CPU multiplier from 20x to 21x to get me closer to my target 4ghz. The turbo boost is always at least at +1x. If only 1 core is in use then the cpu multiplier goes up +2x supposedly but i've never seen it do 2x. Anyways, anybody that says this is not stable simply has not taken the faster clock speed into consideration when setting their voltages.

Currently I have the Uncore running at nearly the same speed as the CPU and I'm running about 1.36v on the processor (supposed to be around 1.38v but it drops a little under full load. I origionally could not get the chip stable past 3.6ghz until I enabled load line calibration.) All other voltage settings are currently auto because no matter how much I played around with the voltages the only two that I have seen make a difference are the CPU and MEM volts. Increasing the Uncore speed has increased my Folding at home clients points per day from 4100 to 5100

 

[Brolloks]

200 x 20 @1.40 volts after 11.5 hours I started using it while Prime95 was running and it rebooted.

200 x 20 @1.45 volts after a few hours I started using it while Prime95 was running and it rebooted.


so now I'm not gonna touch it while Prime95 is running but I switched to 211 x 19 = 4009 MHz @ 1.50 volts.


I booted with BCLK at 220 and ran Memtest for a bit without errors, I guess I should have found out my highest BCLK first but what exactly is the advantage of having higher BCLK at same MHz speeds?

Is Higher BCLK worth a loss of a few MHZ and if so where's the break even point?

Your results are pretty good IMO, expecially getting 220 bclck as very few i7's I've had and seen could do above say 210, even 200...I had two i7-920 that could not go above 200 irespective of voltages.

Seeing that you have flexibility in your QPI, Uncore and RAm speeds using various multis, go for a stable bclck and then push the other speeds.

In general there appears to be a lot of variety in the C0 chips, the new revision will have a more uniform platform I suspect.

As far as boards are concerned they are pretty much the same, bios revisions thus far do make a difference as expected.

Heat is not really and issue, although they all run hot throttling takes place at a much higher temp than 45nm LGA 775 quads.

Stability is normally impacted by v-core only, if you keep RAm speeds low then you should not have to puch Vtt (QPI/VDIMM) voltages above 1.4. The rest is not that impactfull unless you run SLI or quad SLI the IOH needs some v-juice.

Selection of RAM and board compatibility is key to maintain low stable latency (CL7) at speeds around 1600 which seems to be an optimized level, above 1600 you sacrafice latency and performance from my experience.

Running extended high load programs takes its toll on stability, I have yet to get a 12 hr stable setting on air/water above 4.2 Ghz (200 x 21).

Lasty if you use a Gigabyte board use Ntune 6 to finetune your settings in windows if you use turbo mode, booting up especially in Vista sometimes crosses the stability threshold, so just get into Windows and then tweak.

In general there is still much to be learned about the i7's, it is a very interesting platform with much potential...as it is still new variation makes it hard to standardize settings and to optimize OC levels.

 

[c627627]

Thanks. I bought my i7 on Day 0 (Sunday) as soon as it was released, it's from the very first retail batches. I am now in the 12th hour of Prime95 Torture Test at 211 x 19 = 4009 MHz @ 1.50 volts.

I did not touch the computer this time which begs the question what would have happened if I did or if i didn't touch it while testing under lower voltages. Case temp is 36°, the CPU is at 87° Celcius (under Torture stress).

I plan on giving it a few more hours of Prime then using FurMark: http://www.ozone3d.net/benchmarks/fur/
to do a stress test video. I haven't installed it yet so I'm not for sure how long it should be run to confirm stability?



If I am stable, I need help deciding what to tweak next from the current 211 x 19 = 4009 MHz @ 1.50 volts:

1. If I drop to 18: then I would need 18 x 223 to get 4014 MHz which is unrealistic plus I should have some room I think.

2. If I stay with 19: I may get just a little higher than 211, but again leaving some room may be better?

3. I take it you all agree going up to 20 is not recommended since I would have to drop from 211 down toward 200... (I'm still trying to get the definition of why higher BCLK at same MHz is better.)


So that leaves reducing voltages and/or tweaking timings on 3 x 1GB G.SKIL DDR3 1333 (PC3 10666) RAM.


Please advise on what you would do next if you were stable at:

CPU Ratio Setting [19]
BCLK Frequency [211]

CPU Voltage [1.50000]
CPU PLL Voltage [2.00]
QPI/DRAM Core Voltage [1.35000]

IOH Voltage [1.26]
ICH Voltage [Auto]

DRAM Bus Voltage [1.64]

CPU Differential Amplitude [700mV]

CPU Clock Skew [Delay 100ps]
IOH Clock Skew [Delay 100ps]


RAM TIMINGS? I have no clue about where to start with DDR3, help on timings would really be appreciated.

 

[Deanzo]

211 is the max stable BCLK's (if talking the likes of Prime) I've ever had.
For some benchmarks I can do 117, and upto 224 max for just a screenshot.

That your 211 stable now, is very good imo.

My first 940 needed lots of QPI/VTT to get up and running, like 1.55+

It does take a little playing around to find the right spot for your cpu,

Some have found "20" an issue when it comes to clocking a 920, try "21" (turbo on) and clock up in windoz to see if it's any better. I'm not saying it will, but just try it. With the 940 "22" clocks very poor for me, I can get higher speeds with "21"

Added PLL seemed to help high BCLK's on my Giga board, but going over stock (1.8) on the eVGA seemed to do nothing, in fact going over 1.96 seemed like it was holding me back.

eva2000 (a member here, but more so XS) found playing with IOH/ICH added BCLK's for him, though I've had no luck with this as yet. But again do try it out.

 

[c627627]

I'm about to stop the Torture Test now. Thank you for your input, I will not try to push it further as I think it should be close to its limits, it's not about every last MHz for me, I would only do it if I think I could get at least 3%-5% more which is unrealistic.

Instead I could try to reduce voltages and other settings but then I guess since I would have to wait half a day again to confirm stability, it's important I make good educated tries, I'm thinking of testing 211 x 19 = 4009 @:

CPU Voltage [1.50000] reduce to 1.45
CPU PLL Voltage [2.00] reduce to 1.96
QPI/DRAM Core Voltage [1.35000] keep the same

IOH Voltage [1.26] ??

DRAM Bus Voltage [1.64] Reduce but to what?

CPU Differential Amplitude [700mV] Reduce to Auto

CPU Clock Skew [Delay 100ps] Reduce to Auto
IOH Clock Skew [Delay 100ps] Reduce to Auto


What would you guys do?


P.S. While I was originally testing @1.4 volts, I made a grave mistake of installing my power supply with the fan facing down even though I just changed the case to Antec 900 which has the PSU mount at the bottom, instead of the top, I don't know if that contributed to previous failures but facing the PSU fan up and increasing the voltage to 1.50 did the trick, although I did also drop the multi to 19...

 

[c627627]

Wow, over 13 hours of Prime95 eight thread Torture Test stability but FurMark took the system out, BSOD maybe 7 or 8 minutes into it, the temperatures displayed I assume are for the graphic card? They climbed and surpassed the 106° mark last I looked. (Trying to find any log files...)

I have an ATI Radeon HD4850, the Asus version with better cooling, still I heard they are notorious for running hot. So now I really need help. Is this graphic card temp related or what?


I did FurMark > Stability Test > Fullscreen > 1680x1050 (my lcd res) > I did not check Xtreme Burning Mode.

 

[Brolloks]

As Dean said, priming at anything above 210 normally results in a core failure, BSOD or the likes

 

[c627627]

I am very much stable at 211 x 19 = 4009 after 13 hours of Prime, I am looking to repeat that stability at lower voltages/other settings and was looking for help on educated guesses on how low I should go with voltages since I'll be wasting half a day on testing...


The BSOD was a result of FurMark graphic stress test, after which I replicated the BSOD at stock settings thereby confirming :

I would not advise running Furmark, it's pure hell on hardware. The level of stress given will never be matched by any other graphics application.

So no to FurMark stability tests as a testament of a stable overclock. But I am still looking for some other kind of graphics stability test...


These are my educated guesses so far:

CPU Voltage [1.50000] reduce to 1.45
CPU PLL Voltage [2.00] reduce to 1.96
QPI/DRAM Core Voltage [1.35000] keep the same

IOH Voltage [1.26] ??

DRAM Bus Voltage [1.64] Reduce but to what?

CPU Differential Amplitude [700mV] Reduce to Auto

CPU Clock Skew [Delay 100ps] Reduce to Auto
IOH Clock Skew [Delay 100ps] Reduce to Auto

 

[Brolloks]

Actually FUR stress testing is a good combo to test both the GPU and CPU , I use it often to test rigs for gaming stability, if it passes that I don'ot bother with Prime as games do not fail often due to CPU stability but more to GPU stability.

Leave PLL on auto, no need to have IOH above auto unless running SLI

VDIMM at 1.65 is low enough, even at 1.75 there is no harm, as long as the variance between QPI/VDIMM and VDIMM does not exceed 0.5v you'll have no risk in affecting the CPU adversely

 

[jason4207]

I haven't played w/ I7 yet, but I like to get rock-solid stable.

I use P95 large-fft, memtest, & OCCT2.0 custom 2hr RAM test to stress the NB and RAM.

P95 small-fft to stress the CPU.

Fur and ATI Tool to stress gfx, but I run both in windowed mode (FUR in full-screen is too much IMO) at the same time so I can see the artifacts in ATI Tool easier. ATI Tool also has an audible 'dong' if there is a large enough error that I can hear from the other room. Works nicely if I need to do something in the other room.

Oh, and I always put all my fans, especially gfx card, at full speed during stress tests.

 

[Lukee]

I am very much stable at 211 x 19 = 4009 after 13 hours of Prime, I am looking to repeat that stability at lower voltages/other settings and was looking for help on educated guesses on how low I should go with voltages since I'll be wasting half a day on testing...


The BSOD was a result of FurMark graphic stress test, after which I replicated the BSOD at stock settings thereby confirming :

So no to FurMark stability tests as a testament of a stable overclock. But I am still looking for some other kind of graphics stability test...


These are my educated guesses so far:

CPU Voltage [1.50000] reduce to 1.45
CPU PLL Voltage [2.00] reduce to 1.96
QPI/DRAM Core Voltage [1.35000] keep the same

IOH Voltage [1.26] ??

DRAM Bus Voltage [1.64] Reduce but to what?

CPU Differential Amplitude [700mV] Reduce to Auto

CPU Clock Skew [Delay 100ps] Reduce to Auto
IOH Clock Skew [Delay 100ps] Reduce to Auto

Wow, that Vcore is pretty high. I'm running 211 x 19 HT On @ 1.296(bios), 1.288 CPU-z

I suspect that you can go down to 1.35 or so and start from there.

 

[c627627]

Yeah I passed at 1.45v and setting many other settings to Auto. Next up I'll try 1.4 again.

My video card can't pass FurMark at stock because of the known inadequate cooling problem for the Ati 4850 cards:
http://www.ocforums.com/showthread.php?t=595731


You know I was thinking of trying to increase DRAM Frequency [DDR3-1269MHz] For my 3 x 1GB G.SKIL DDR3 1333 (PC3 10666).

The problem is that the next setting is above 1600 and I can't boot with it, my question is: more voltage or leave it be and try better timings @ [DDR3-1269MHz]

 

[c627627]

I have to post this in this thread: I had a PCI sound card installed directly next to the video card, effectively blocking air circulation into the video card, after taking it out, max temperatures dropped from 108° crash temps to 76° C, easily passing the recommended 15 minutes of Furmark.

So next time you encounter this, check if the problem is a PCI card installed next to the video card.


What is the downside of installing the video card in the other slot, then installing the PCI card above it?

 

[c627627]

Final overclock is
Intel i7 920 [211] BCLK x 19 = 4009 MHz @ 1.4 CPU Voltage
3 x 1GB G.SKIL DDR3 1333 (PC3 10666)[DDR3-1691MHz] 10-10-10-24 @ 1.64 DRAM Bus Voltage
ASUS P6T Deluxe v.1 [LGA 1366 Intel X58] BIOS 1102
Thermalright Ultra-120 eXtreme 1366 RT with 120mm Scythe S-Flex F fan
ASUS EAH4850 TOP Radeon HD 4850 512MB @ 680 MHz GPU & 2100 MHz Memory
Antec nine hundred case, two front 120mm fans, one back 120mm Fan, one top 200mm fan
Corsair CMPSU-750TX 750W


The video card is only stable at extreme FurMark temperatures if there is no PCI card next to it, blocking air flow. Switching the video card to the other slot doesn't help because my PSU is bottom mounted, again blocking air circulation. Therefore, getting rid of the PCI card next to it, is the only solution for video card stability at maximum stress.

 

[Brolloks]

Looking good , 4 Ghz is the sweet spot for these i7's, anything above takes either a lot of cooling, v-core ... in short a lot of work and heartburn, there are the exceptions off course as always.