Can't figure out why i7 920 not stable at 4GHz...

[r2rX]

I hope so too, xrror.

This is what is weird. The following settings are what i'm testing at the moment:

BCLK = 192
Multiplier = 21 (EIST enabled)
Memory Ratio = 5 (to get DRAM 1920MHz)
Uncore Ratio = 10 (to get 3840MHz ; double the RAM value)
QPI Frequency = 4.800GT (resulting in QPI Frequency @ 6912MHz)
QPI Voltage = Auto (don't have a program to determine what it's set to but if I follow MSi Control Center it's 1.59)
PCI-E Frequency = 100
DRAM Voltage = 1.65
CPU Voltage = 1.336 (will work on trimming this a little more if possible)

With these settings my system is rock solid. Prime95 churns with no problem, LinX processes max memory available with no freezing or stuttering at all. Not to mention that games are running fine and no crashing whatsoever.

It's an odd feeling at this point....one of the main reasons I started investigating my systems instability was due to the fact that I would get frequent crashes with the games that I play. Since determing the 200x20 with QPI @ 1.65, the very same games do not crash anymore. The only thing bugging me is that Prime95 and LinX crashes; you know, that feeling where you know that the system isn't 100% stable even though it works fine for what you need.

Now I know that media encoding, 3D work and gaming are not gonna push the system anywhere near the extent that LinX and Prime95 do, so i'm contemplating if it is worth really worrying about aside from the fact that the QPI is stupidly high (cause after all my searching on the net i've not seen anyone needing to put their QPI on 1.65) and I don't want to jeopardize my system/CPU if having the QPI so high is an obvious way to do that.

Furthermore, with the above settings which i'm testing, when I reach 193 I start getting few errors within LinX and Prime95....the higher I raise the BCLK the more errors start popping up until I reach 200 where the moment there is an error (in either of them) my system crashes.

So is the BCLK / Uncore at the max for this particular CPU? Or could it be the board? I'm not sure at this point.

But I appreciate all the feedback so far, guys. If anything, i've learned quite a lot about i7's and overclocking....a little different from OC'ing Core 2 Quads.

Still willing to experiment.

EDIT: Another question; with EIST enabled (where C1E and other speed stepping features are disabled) I get an extra core unlocked. So Windows registers 3840MHz but CPU-Z and other diagnostic programs recognize 4032MHz. Now, is my system running at the latter or are there specific conditions before it does? And if so, does it apply to all cores or just the first core?

Again, thanks.

r2rX

 

[redduc900]

Consider the absolute max. VTT is 1.45V, although I have seen as high as ~1.59V on some setups, and I've recommended as high as 1.5-155V myself. I doubt anyone actually runs w/ it that high long term, as most just lower the DRAM frequency and in turn lower the VTT to a more reasonable level. Also most folks that purchase high speed DRAM like DDR3-1866 to DDR3-2133 don't realize that VTT for the most part needs to be increased well above what Intel recommends as the max. for a stable system, of course dependant on the quality of the IMC and the specific board in general.

 

[djscrew]

Hey guys.

To answer some of the questions, the following are my settings:

BCLK = 200
Multiplier = 20
Memory Ratio = 5 (to get DRAM 2000MHz)
Uncore Ratio = 10 (to get 4000MHz ; double the RAM value)
QPI Frequency = 4.800GT (resulting in QPI Frequency @ 7200MHz)
PCI-E Frequency = 100 ( no difference @ 110, but safer on 100)
DRAM Voltage = 1.65 (tested on 1.7 but results the same)
CPU Voltage = 1.350 (with droop it is 1.344)

These are the base settings. The CPU Voltage might seem a bit high but i've followed the patterns that i've seen when the CPU Vcore is set to 'Auto'. On Auto, the CPU Vcore is 1.31v. When running LinX (for example), it jumps from 1.31 to 1.34...and then BSODs with the error 0x101 (which implies increasing Vcore). So I manually set it to 1.34v and, when being stress-tested, increases to 1.36v....with this, specifically, it does not BSOD with a Vcore-specific error. I've tested putting the Vcore lower than 1.31, but it is not stable....so those who can get their 920 D0 @ 4GHz with less than 1.3v are fortunate.

The following are the RAM timings that i've experimented with: 8-9-8-24-(1T and 2T tested), 9-9-9-24-(1T and 2T tested) and 10-10-10-30-(1T and 2T tested). I've not seen any difference, in relation to stability, using any of the above timings.

Now, the following is the progress i've experienced so far......

Part of the issue I was having was definitely QPI related. My BIOS limited the max value to 1.52. However, I remembered that there is a switch on my board that increases the cap value in my BIOS. So, once switched, the max QPI value went to 1.75.

So, I started testing...and, basically, I managed to get something semi-stable..ish.

VCore = 1.350v
QPI = 1.650v

I spent a couple of hours testing the values between these two. The higher the Vcore, the lower the QPI can be (i.e. VCore @ 1.350 is "stable" with QPI @ 1.650 but Vcore @ 1.375 is not stable with the same QPI etc). I tested from low Vcore and low QPI and increased QPI. Then increased Vcore a little and retested with low QPI and then increasing QPI etc.

So, basically, the closest to the "sweet spot" that i've discovered so far is Vcore @ 1.350 and QPI @ 1.650. This configuration is the only one where i've managed to pass a couple of LinX tests without crashing immediately...although it will only pass with 512MB selected...and will eventually crash after 5-10 passes. Increasing the mem size beyond 512 results in immediate crashing. Prime95 crashes but not immediately.

Irrespective of CPU Vcore values i've tested, the QPI will not be stable at 1.35 or below. So, if I set the QPI lower than 1.65, I either get booting issues (which occur with low values i.e. 1.4 and less) or instability/crashing when stress testing. If I set the QPI higher than 1.65 my computer turns off when stress-testing.

The hours of testing included trying out different RAM timings (and voltage), testing with PLL and IOH values.

The end results is that when the CPU Vcore is at 1.350v and the QPI is at 1.65, the results are the same irrespective of PLL, IOH or RAM (timing and voltage) settings.

Concerning read-outs in AIDA64, when selecting 'Motherboard' -> BIOS , it should list BIOS Properties, BIOS Settings and BIOS Manufacturer; the BIOS Settings being the displayed info CPU PLL, Voltage, QPI etc. On my system, the BIOS Settings are not present; only BIOS Properties and BIOS Manufacturer....so I can't use AIDA64 to monitor those values while stress-testing.

The only other option is MSi Control Center...which is showing some strange results. It doesn't detect the correct QPI value when I set it manually and when left to Auto (in the BIOS) it picks it up as 1.20...and then changes to 1.59. MSi Control Center also detects IOH @ 1.45.....and then changes to 1.10....and the CPU PLL @ 1.95.

And one more update: I enabled the OC-Genie button on the motherboard, then went into the BIOS and checked it out. The CPU and RAM were clocked slightly lower than my settings...but it auto-adjusted 4 settings: VCore @ 1.4, QPI @ 1.51, IOH PCI-E @ 1.548 and DRAM Voltage @ 1.65v........those were the only settings adjusted; everything else was left on AUTO. I replicated these settings on my clocks but it does not work (i'd come across the same combination during my hours of testing prior).

...so what do you guys think?

r2rX

sounds like you've been taking the "right" (albeit risky) approach to this with your methodical testing, here's what I think:

You're pushing your IMC too hard, it simply cant handle the 7.2 ghz QPI + the ram clock speed. You have done all of the right moves with voltages and gone much farther then I would have in testing to see if voltage would get you where you want to be. You're just plain out of room for the QPI/DRAM bus to improve any further. This is evident when you test over 512 megs of Ram, the additional power draw on the ram causes your QPI voltage to plummet and cause the system to crash, I have the same problem with mine. I just wasn't quite able to get it to 8-8-8-20-1T like I wanted

If you want to OC your ram or even get it to stock, you need to lower your BCLK. Alternatively you can buy an Extreme Edition chip with fully unlocked QPI multipliers for ~$1k which would let you easily lower the QPI 1 multi at a time.

I must commend you for the massive size of your nuts in pursuing the best overclock possible

EDIT: its possible that 1.5v ram at the same speeds would give you more room, see CMZ6GX3M3A2000C10 on the egg.

 

[Woomack]

Consider the absolute max. VTT is 1.45V, although I have seen as high as ~1.59V on some setups, and I've recommended as high as 1.5-155V myself. I doubt anyone actually runs w/ it that high long term, as most just lower the DRAM frequency and in turn lower the VTT to a more reasonable level. Also most folks that purchase high speed DRAM like DDR3-1866 to DDR3-2133 don't realize that VTT for the most part needs to be increased well above what Intel recommends as the max. for a stable system, of course dependant on the quality of the IMC and the specific board in general.

I will agree here. I couldn't make 2000+ stable under ~1.55V QPI/VTT on at least 3 boards and 3-4 cpus. Usually ASUS boards need a bit higher voltage but still 2000+ in 95% cases = 1.55V+VTT, that isn't best idea for 24/7 on Bloomfields.

@r2rX , If you can make it work @1.65V qpi and it's crashing then probably something is overheating and it won't run stable enough to pass longer full load tests.
If you compare performance then 1600 6-8-6 will be about the same as 2000 8-9-8 using GTX3 memory. It's one of reasons why most users just pass anything 1800+ and try to stick to 1600 on tighter timings.
If you used all tips in this thread then probably you checked everything.

EDIT: its possible that 1.5v ram at the same speeds would give you more room, see CMZ6GX3M3A2000C10 on the egg.

There is no way that Vengeance kit will beat GTX3 in anything Dominators GT / PSC are making up to 2500MHz on tighter timings than Vengeance can make @2000 Voltage won't change anything in this case.

 

[r2rX]

Thanks for the feedback, guys.

I think I understand this situation much better now. Unfortunately I cannot buy anything else at the moment (already forked out quite a bit...and a GTX680 is on the way, so i'm broke as a joke ).

It seems I reached the limit of what my system can handle. The problem is that when the QPI reaches 7GHz (and above) my chip/board just doesn't like it. By setting my chip to 200x20 it will inevitably become unstable....I already know this as I tested my RAM on 1200 and 1600 (with different timings + voltage) only to get the same results.

So, from all the time spent testing different configurations, the following is my maxed out settings:

BCLK = 192
Multiplier = 21 (EIST set to Auto for extra multiplier)
CPU PLL = Auto
IOH = Auto
Memory Ratio = 5 (to get DRAM 1920MHz)
Uncore Ratio = 10 (to get 3840MHz ; double the RAM value)
QPI Frequency = 4.800GT (resulting QPI Frequency @ 6912MHz)
QPI Voltage = Auto
PCI-E Frequency = 100
DRAM Voltage = 1.65
DRAM Timings = 8 : 9 : 8: 24 : 1T
CPU Voltage = 1.320 (This is as low as possible before becoming unstable)

So I think it's pretty good considering. There was alot of information I didn't really know/appreciate before buying this hardware, so all you guys have helped me learn quite a bit. A friend donated his old i7 920 chip and Zalman cooling kit so I purchased the Mobo and RAM thinking it'd be simple setting everything to max....

You're pushing your IMC too hard, it simply cant handle the 7.2 ghz QPI + the ram clock speed. You have done all of the right moves with voltages and gone much farther then I would have in testing to see if voltage would get you where you want to be. You're just plain out of room for the QPI/DRAM bus to improve any further. This is evident when you test over 512 megs of Ram, the additional power draw on the ram causes your QPI voltage to plummet and cause the system to crash, I have the same problem with mine. I just wasn't quite able to get it to 8-8-8-20-1T like I wanted

I agree that this is the case. My QPI Freq, at the moment, is at 6912. This is literally the peek that the CPU will be stable. Adding one addition point to the BCLK brings the Freq to 6948 and things start to become unstable....and it doesn't matter what RAM speed or timing I use.

I must commend you for the massive size of your nuts in pursuing the best overclock possible

Hehehe Thank you. I love my hardware so i've always pushed it to the best it can go. But you guys helped me alot, so my thanks to you.

I think my CPU @ 3840MHz (or 4032MHz..whenever it kicks in) and my RAM at 1920MHz is pretty good as it is....plus the timings are quite trimmed...so it's snappy; but most importantly it's stable now.

Even though my gaming is fine when the system is set to 200x20 @ QPI 1.65, i'm not comfortable with the QPI being set so high...I would like this equipment to last for a while. I'm not sure what the QPI value is on my "stable" setting (will whip out the voltmeter later) but i'm sure it's a bit lower than 1.65. Also, I doubt there is going to be a significant different between 200x20 + RAM 2000MHz and 192x21 + RAM 1920MHz.

I think this wraps things up. Again, thank you guys very much for guiding and teaching me throughout this process. There is someone else who has a similar system who wants to overclock so I am very prepared to assist him now (so you guys have also helped him too! ).

Before concluding, I would like to ask one last question. It's about the EIST.

In the BIOS, if I set EIST to 'Enabled' then I get a Turbo option. If I set the EIST to 'Auto' then I can increase the multiplier to 21 (no Turbo option). Now, the diagnostic tools (i.e. CPU-Z, AIDA64 etc) indicate that the system is set to 192x21 @ 4032MHz. But if I look at the POST screen, when starting my machine, it says 192x21 @ 3840MHz....also Windows / my desktop gadget says it's 3840MHz.

So when does the system actually kick in the extra multiplier and run at 4032MHz? And when it is running, does it apply to the first CPU core or all of them?

Again.. thanks a million, guys.

r2rX

 

[djscrew]

good to see you found an excellent sweet spot for your system, seems like you landed just about where I did with my very similar build...

I believe the turbo is applied when demand is necessitated, and also the temps are below a certain threshold... nvm, see link:

http://www.intel.com/content/www/us...ology/turbo-boost/turbo-boost-technology.html

 

[r2rX]

Alrighty, I guess that's that. Thanks again, guys. The system is running beautifully.

r2rX

 

[UltraTaco]

On mine, it's also 21x, which is on all cores, but cinebench and some other programs don't recognize it. They keep reporting frequency as it would be without turbo(20x).

Isn't turbo-boost sort of like software related that's why it doesn't show on boot screen, but when everything loaded, it shows up? Well, it does on cpuz.

 

[maxfly]

i dont think anyones home Taco. r2rx hasnt been around since 2012...

 

[UltraTaco]

You're right! Last activity may2012.

 

[EarthDog]

Thread necro's FTW!!

Hopefully we will put measure in place to make people a bit more concisous of what they are doing when replying to threads.