i7 920 OCing to 4GHz - Problem between keyboard and chair or just a bad chip?

[Zoeff]

Some might remember an earlier thread I made asking for purchasing advice. Since I bought it I've spent quite a lot on the 'net figuring out what does what and comparing results. However, there are still some things I don't fully understand and I can't seem to figure out if this chip just doesn't like 4GHz or if the problem is me. I'm trying to get the maximum performance out of it without eating electricity or reducing the life span too much (It's on 24/7).

Here's what I have and what I've configured in the bios.

i7 920 D0 with a Zalman CNPS10X Extreme
P6T Deluxe V2 901
3x2GB (OCZ3X1600LV6GK)
8800GTX
X25-M 80GB
620 Watts PSU
Win 7 64bit

Some BIOS settings...
Multiplier = 19x
BCLK = 200
DRAM Frequency = DDR3-1603MHz
UCLK = 3208 MHz
QPI Link Data Rate = 7218MT/s
CPU Voltage: 1.16875
CPU PLL Voltage: 1.80
QPI/DRAM Core Voltage: 1.21250
IOH Voltage: 1.10
ICH Voltage: 1.10
DRAM Bus Voltage: 1.64
RAM Timings: 8-8-8-24 1T
Load-Line Calibration = Enabled
Intel SpeedStep = Enabled
Hyperthreading = Disabled
C1E/C-STATE = Disabled ...I think. Probably disabled.
CPU Clock Skew = 300ps
IOH Clock Skew = 300ps

Been running Prime95 for just over 15 hours now, no errors.

(Validated while Prime95 was running)

CPU-z reports my core voltage as being 1.160 when priming. Quite odd considering this is actually LOWER than the default voltage at much higher speeds. It drops slightly under that when idle. I'm guessing this is because of LLC? Is this a good thing? I'm somewhat confused with LLC as some say that it also increases the peak voltage during transients but would that matter if the steady state voltage is quite low anyway?

The frequency oscillates between 3799.9 and 3800.1 with the occasional spike at 3799.8 and 3800.2 while priming. Does this matter?

CPU Temp while Prime95 is running peaks at 55 C, at idle it's somewhere between 30 and 35 C.

When the voltages are set higher, then physically touching the northbridge/southbridge heatsink (they're connected) or the QPI heatsink for more than 2 second my finger starts to complain. There are no temperature sensors over there so this is all I can gather - Is this normal? Both are designed with a regular CPU cooler in mind that blows air directly towards the CPU and spreads out in all directions. The motherboard did come with a small fan in the packaging which is supposed to be used if a passive cooling system is used. Considering my current cooler blows air sideways and directly out of the case it seems to be a reasonable precaution to install it but there's no more room left over. Should I worry about these temperatures?

What exactly is clock skew? I've changed it from auto to 300ps just to try it but I'm still in the dark...

The QPI voltage is higher than the core voltage, is this normal?

This is why first overclocking attempt of any computers I've owned other than some curiosity driven stabs at overclocking a P4. Yet an i7 seems to be almost made for overclocking, why doesn't Intel just ship them with higher clockspeeds to begin with? Are there situations where these things actually don't overclock too well?

Now the 4GHz problem - Whenever I try to run it at this speed using various combination's such as 19x211, 21x191 or even 20x200 (which some say isn't a good choice) I need to bump the voltage much, much higher. Everything except the core voltage is set much higher and the various features such as SpeedStep are disabled to make sure that any of that isn't the problem yet for a 200MHz increase I'll need to increase the core voltage beyond 1.3 which suggests the problem is somewhere else. Keep in mind that this is my first overclocking tour so I might be missing the blatantly obvious...

If you're wondering why I even bother trying to reach 4GHz with an 8800GTX, I'd like to see how far this particular chip will go and will upgrade my graphics card at some point considering it's the biggest bottleneck. That and I like the technical side of it.

Thanks!

 

[qazwsxokmijn]

1.16v is too low for high clocks. My D0 does 3.8GHz at 1.246v, 4GHz at 1.32v.

 

[SunnyD]

1.16v is too low for high clocks. My D0 does 3.8GHz at 1.246v, 4GHz at 1.32v.

Yeah, pretty much this.

 

[Blastman]

The i7-920 is a Bloomfield core which is made using a 0.45nm process technology. The 0.45nm cores have difficulty making it to 4.0Ghz without some serious overvolting. It’s only the newer 0.32 process technology cores like the i3-530 that hit 4.0GHz quite easily.

xbit …In fact, frequencies around 3.8GHz are the most widely spread maximum for Core i7-920 CPUs overclocked with air coolers. This is the conclusion we can make judging by our test results and basing on the feedback from the first users who purchased these processors. By the way, the engineering sample of the Core i7-920 CPU that we also have in our lab demonstrated the same overclockability. Even with the Vcore increased to 1.4V it overclocked only to 3.8GHz.

That's par for the course. In my view, the best thing to do is back off from the 3.8Ghz and run a little slower -- like the 3.3 - 3.5Ghz with just a minor voltage tweak if needed. That will be a nice jump from the stock speed of 2.66 and will be faster than the 3.2Ghz of the i7-955 Extreme Edition.

 

[Gillbot]

IMHO, your first mistake is expecting to hit 4GHz.

 

[VirtualLarry]

Sounds like 3.8Ghz is your "sweet spot". Stay there, is my opinion.

My Q6600 in sig overclocks to 3.3Ghz with default vcore. IMHO, that's probably the sweet-spot of that chip as far as cooling goes. I had to repeatedly increase the limit for the BIOS temp alarm (to 95C!), to allow for a 3.6Ghz overclock. I figure that's still somewhat safe, since I don't run OCCT:linpack normally, and that can get like 5-10C hotter than any other program.

Really though, the only reason that I pushed on for 3.6Ghz was bragging rights. I don't even use the quad-core for anything, really, it just sits there. My E2140 dual-core is just chugging along fine for me. I run distributed computing on it, and my magicjack.

since it's 65nm, power consumption isn't great, but the Q6600 @ 3.6 is much worse.

 

[Ben90]

CPU-z reports my core voltage as being 1.160 when priming. Quite odd considering this is actually LOWER than the default voltage at much higher speeds. It drops slightly under that when idle. I'm guessing this is because of LLC? Is this a good thing? I'm somewhat confused with LLC as some say that it also increases the peak voltage during transients but would that matter if the steady state voltage is quite low anyway?
Microprocessor companies use slightly higher voltages than what is needed. Its much easier for them to raise the voltage .1 than it is for them to go through strenuous testing to find the absolute limits of the chip. Also having HTT turned off helps you keep a lower voltage.

No one (except the engineers) quite knows the implications of having LLC turned on vs off at high voltages. There are theories for both sides, but the truth of the matter is it really doesn't matter until your at or slightly above 1.4v.

The frequency oscillates between 3799.9 and 3800.1 with the occasional spike at 3799.8 and 3800.2 while priming. Does this matter?
The oscillations come from a PLL. These distribute clock timings to the CPU, but are never 100% accurate. This is perfectly normal.

When the voltages are set higher, then physically touching the northbridge/southbridge heatsink (they're connected) or the QPI heatsink for more than 2 second my finger starts to complain. There are no temperature sensors over there so this is all I can gather - Is this normal? Both are designed with a regular CPU cooler in mind that blows air directly towards the CPU and spreads out in all directions. The motherboard did come with a small fan in the packaging which is supposed to be used if a passive cooling system is used. Considering my current cooler blows air sideways and directly out of the case it seems to be a reasonable precaution to install it but there's no more room left over. Should I worry about these temperatures?
The stock Intel cooler helps out everything from the Northbridge to the VRMs on a Mobo and even system RAM. Its a tough call since I can't touch your NB physically, but most likely things are still okay. If you are not that comfortable with the temps then you might want to look into ghetto rigging a small fan onto it using a rubber band or something.

The QPI voltage is higher than the core voltage, is this normal?
Yea that is fine.

This is why first overclocking attempt of any computers I've owned other than some curiosity driven stabs at overclocking a P4. Yet an i7 seems to be almost made for overclocking, why doesn't Intel just ship them with higher clockspeeds to begin with? Are there situations where these things actually don't overclock too well?
Intel really has no need to ship higher clock speeds currently. They control the high-end market, and having faster processors out now means less of a reason to upgrade later. Also, these processors run very hot under load and it may be very tough for them to keep the TDP within limits. Remember the i7-975 (and i7-980?) boost up to 3.6Ghz already so there isn't that much margin to increase speeds and maintain server level reliability.

Now the 4GHz problem - Whenever I try to run it at this speed using various combination's such as 19x211, 21x191 or even 20x200 (which some say isn't a good choice) I need to bump the voltage much, much higher. Everything except the core voltage is set much higher and the various features such as SpeedStep are disabled to make sure that any of that isn't the problem yet for a 200MHz increase I'll need to increase the core voltage beyond 1.3 which suggests the problem is somewhere else. Keep in mind that this is my first overclocking tour so I might be missing the blatantly obvious...
Every chip/system overclocks differently. Since you seem stable at a 200 Bclock, I would try the 21x191 so we know any stability problems lay with the processor. If you still need 1.3v to get that stable then we know its your processor. Your processor seems to love 3.8Ghz as your using a very low voltage, but 4Ghz may be unreachable with a voltage you desire. I was really set on breaking the 4Ghz wall, and I CAN get my processor up to 4.2, but I just needed too much voltage for me to run comfortably since I'm on the stock cooler and summer is coming so I dropped it down to 3.8. 4Ghz @ 1.3v is not unreasonable so don't feel like you have a dud. It just may not be a ruby-chip.

My responses in bold

 

[OILFIELDTRASH]

Looks like you should give your qpi a little voltage, 1.3-1.34, intel spec is 1.35 max. That chip looks to be a good one. 3.8ghz on 1.18v is great. You should hit 4ghz easy if you have the cooling to do so. Little voltage to the cpu and qpi and you are good.

 

[Zoeff]

Thanks for the responses so far, though it seems the config stated in the original post isn't all that stable after all. At roughly the 16 hour mark of Prime95 blending my computer suddenly froze. Not even a BSOD.

Meh, I guess that this means it's stable...just not Prime95 stable ~16 hours in. I guess that's reasonable.

I'll keep it at 3.8GHz with 1.16v then.

@Gillbot: Everybody seems to be able to hit 4GHz from what I can tell, just not all at the same voltages.

 

[DrMrLordX]

A few points.

1). 3.8 ghz on a 920 is pretty good stuff. If you can tune up your memory and uncore speeds then you are looking at some serious performance potential. Hell you're looking at serious performance even if you leave your memory and uncore at less-than-aggressive settings.

2). If your board is heating up under various sinks then you need better sinks and/or better airflow. The more voltage you throw at pretty much any part of your system, the worse that situation is going to get (you really need to keep your VRMs cool, they could go poof if they get hot enough for long enough).

I don't know what case you're using, but one cheap-ass trick I've found is to pull the side off the case and to blow a fan into the side of the computer.

On my last build, I tried this with a massive 2200 cfm room fan, which helped some but made a lot of noise (just from the air moving) and moved air into places where it didn't necessarily need to go. It was a big fan, and most of the air got moved by the edges of the fins, which in turn wound up slamming into the case perimiter/frame instead of directly inside the case. My attempts to build a cone to force the air onto a smaller target failed due to lack of crafts skills and the realization that my room fan generated insufficient static pressure for such a scheme to work.

On this build, I bought a $15 Honeywell room air circulator, which is nothing more than a Honeywell knock-off of a Vornado. It might produce some notable static pressure, but I'm not going to do anything to try and find out. The only important thing about it being a Vornado knock-off is that it does a good job of focusing airflow to a small(er) area as compared to the massive room fan I used, even from relatively long range. I use mine to keep air moving onto my RAM and to provide a perpendicular feed for my nh-d14, though you could easily use one to cool hotspots on your motherboard elsewhere. It doesn't make much noise, even on the highest setting, and it cost me $25 less than would a real Vornado.

I must admit that blowing a room fan into an open case is pretty ghetto/trailerpark/whatteveryouwannacallit, but it can work. Enjoy the dust!

 

[Zoeff]

VRMs, those are the things underneath the heatsink that's next to the CPU socket, right? That one does get somewhat toasty. I do have a room fan actually and my case is open anyway. Might try that...

 

[Ben90]

Lol DrMrLordX, I do quite the opposite. Instead of having a room fan blowing to cool my computer. I have a 120mm fan plugged into the mobo keeping me cool. It also keeps me entertained when I have nothing else to play with.

 

[Mr. Pedantic]

The frequency oscillates between 3799.9 and 3800.1 with the occasional spike at 3799.8 and 3800.2 while priming. Does this matter?

Can you tell the difference between the 200kHz? If you can't, then I wouldn't worry about it. If you can, too bad. There's nothing you can do about it anyway.

The more pertinent question is, can you tell the difference between 3.6 and 3.8GHz? If you can't really tell a difference between these either, then I wouldn't worry about trying to get 4GHz, unless you're doing it for your e-peen.

 

[DrMrLordX]

VRMs, those are the things underneath the heatsink that's next to the CPU socket, right? That one does get somewhat toasty. I do have a room fan actually and my case is open anyway. Might try that...

The VRMs could be in different locations depending on the motherboard. On any board that isn't cheap as hell, they will be under a sink somewhere.

VRM temperatures get very high when you start pushing a lot of power across the board, and while it varies from board to board and platform to platform, temps up to and in excess of 90C are not unheard of. You don't want them to get that hot though.

I think it's fairly safe to assume that whichever heatsink on your mobo is getting the hottest is cooling VRMs.

edit: I meant PWMs, not VRMs. Dammit! Preserved the rest of my posts to record how much of an idjit I am. Not a huge mis-use of terminology but whatever.

One thing to remember with a room fan: those electric motors probably aren't shielded terribly well. Putting one too close to your system might do something bad. I haven't run into any trouble yet but you never know. The larger the fan, the larger the potential problem, so going with something small like the Honeywell fan I'm using will likely pose no real danger.

Lol DrMrLordX, I do quite the opposite. Instead of having a room fan blowing to cool my computer. I have a 120mm fan plugged into the mobo keeping me cool. It also keeps me entertained when I have nothing else to play with.

Ha! Now that's interesting. Are you using a delta or something like that? whooosh!

On a side note, I used to use my old 1.4 ghz tbird system as a space-heater in the winter. You could feel the waves of heat coming out of it when the side-panel was removed (which was more-or-less a requirement for running that thing near the end of its useful life). That CPU pushed 72-74C on a regular basis.

 

[BTRY B 529th FA BN]

If you vcore is dropping like that, personally I think that would be normal, but on that scope there might be too much pressure on the PSU. From what I gathered, all PSUs power will droop a little bit when the max output goes up. It's an almost certianty to see a graphical explanation like that every time you read a PSU review where they raise the peak power during each test along a curve to the max. The higher towards the max, the more the droop. Read some reviews on JonnyGURU.com to find out more.

 

[n7]

LinX finds errors/instability that P95 doesn't, or at least finds it more quickly, so personally, i wouldn't spend much time bothering with P95.

I've found the above to definitively be the case for my i5 750/i7 860/i7 930.


You're going to need more vcore, possibly a lot more.
And possibly higher vtt/qpi voltage, also.

 

[BTRY B 529th FA BN]

I didn't even catch that. Yes, Linx > Prime95 & Memtest HCI > Memtest86

 

[BlackDragon24]

The reason your voltage is lower than "stock voltage" is because you are on a lower cpu multi. When you lower the multiplier, the "auto" voltages readjust accordingly. You will have the highest auto voltage clock for clock with the 21 multi. I've owned X58 boards and all of them have done this.

Just for reference sake, my 920 needs 1.232v 1.2375 QPI for 4ghz (20 x 200), 1.272v 1.2625 QPI for 4.2ghz (21 x 200), and 1.312v 1.300 QPI for 4.3ghz (21 x 205). And it is considered to be a pretty good batch number. The freezing you see in prime95 after 16 hours most likely can be solved with a slight bump in QPI/VTT voltage. If it were a vcore problem it most likely would have errored out or BSOD'd.

 

[SanDiegoPC]

IMHO, your first mistake is expecting to hit 4GHz.

Bingo. You may be able to 'expect' 3.8GHz out of an i7 920 but not 4.0 - that' unreasonable. If yo can pull it off, great but dont' expect it.

 

[HarryO]

IMHO, your first mistake is expecting to hit 4GHz.

I've overclocked half a dozen 920's (CO's & DO's) and i've never had a problem reaching 4GHz on any of them. That's not to say some CPU's just don't have it in them, but as long as you have proper cooling and a good motherboard 4GHz is not that hard.
The only time I couldn't reach 4GHz was with a crappy Asrock Extreme mobo. The same CPU in a Gigabyte UD3R did 4GHz with eas.

 

[Zoeff]

I've upped the QPI voltage a notch, hopefully no freezes will occur during actual use.

As for LinX, I might try it at some point but for now p95 should suffice.

 

[aigomorla]

Zoeff..

when we overclock...

Typically we set the voltages to max acceptable.

So for you

CPU Voltage: 1.325
CPU PLL Voltage: 1.86
CPU VTT: 1.325
QPI/DRAM Core Voltage: 1.325
IOH Voltage: 1.25
ICH Voltage: 1.25
DRAM Bus Voltage: 1.64

Then you set the OC u want.

As you pass test, you slowly notch your voltage down in all area's by 1 or 2, until it fails.

Then you find out what voltage made it fail, and lock it, while lowering the others.

This is what you call fine tuning.

The way your doing it is shooting for fish in a barrel.
Meaning.. unless u know your chip inside and out... you can run into a LOT of headaches by doing it your way.

Starting low, and then clocking, will not get u want you want.. it will get you best possible.
To get what you want, you need to start high, and then go down low.

 

[Zoeff]

aigomorla, that's what I did and that's how I found out about the high voltage needed for 4GHz compared to 3.8GHz.

 

[OILFIELDTRASH]

Zoeff..

when we overclock...

Typically we set the voltages to max acceptable.

So for you

CPU Voltage: 1.325
CPU PLL Voltage: 1.86
CPU VTT: 1.325
QPI/DRAM Core Voltage: 1.325
IOH Voltage: 1.25
ICH Voltage: 1.25
DRAM Bus Voltage: 1.64

Then you set the OC u want.

As you pass test, you slowly notch your voltage down in all area's by 1 or 2, until it fails.

Then you find out what voltage made it fail, and lock it, while lowering the others.

This is what you call fine tuning.

The way your doing it is shooting for fish in a barrel.
Meaning.. unless u know your chip inside and out... you can run into a LOT of headaches by doing it your way.

Starting low, and then clocking, will not get u want you want.. it will get you best possible.
To get what you want, you need to start high, and then go down low.

What are you using to stability test it quickly? IBT? When I see someone using linx are they referring to IBT?

 

[Gillbot]

@Gillbot: Everybody seems to be able to hit 4GHz from what I can tell, just not all at the same voltages.

Do a search in here and many other forums and tell that to all the people who aren't hitting 4GHz. I would bet there are far more that aren't getting 4GHz as opposed to the ones that do and post to brag about it.