Core i7/X58 Overclocking Thread

[Lukee]

1.79 qpi/vtt

way too high it must be something else

I don't think so. I've lowered CPU/Uncore/Memory multipliers and just raised Bclk.

I did 200 x 19 ok, then in order to boot into windows and not lock up during Prime small fft at 205 I had to raise qpi/vtt to over 1.65., then 1.79 for 210.

I guess I can try it again, but I would get it either not posting at all, or hanging at the windows splash screen.

---

Tried it again and it just keeps restarting. When I fed it a lot more Qpi/vtt then it would boot up. Oh well. I mean, I can do 4.2 with HT on so it's not a totally bad chip.

[ReverendMaynard]

I for one am pretty sick of this Prime95 holy grail .. OK so I cannot get more than a couple of hours @ 4Ghz of Prime to save my life with reasonable voltage ..

So what ???

I have my 'game stable' setup which involves the following:

- At least 5 mins of OCCT
- At least 5 mins of Arena
- At least 5 loops of Cinebench
- At least 5 loops of wPrime
- At least 5 mins of Everest Stability
- At least 5 loops of SuperPI

- At least 2 hours of Crysis/Assassins/Far Cry 2/POP etc ...

This may sound pretty lame for an OC test .. but I have pretty much proven if it passes these tests.. it will play games for hours .. and encode DVD's etc. . and thats the main reason I bought the machine ..

Besides I've also heard enough stories of machines that are prime stable for 15 hours then fall over idling or playing simple games ..

I know which machine I'd rather have .. game stable not prime stable

That doesn't sound lame at all. I've always, going back 10 years of ocing, have been a firm believer and advocate of real world stability. Running prime for 12 - 24 hours is just shaving off the top end of your systems lifespan all for the sake of proving something that you shouldn't have to. I've found, especially here, that some dudes get seriously bent out of shape when people refuse to meet their stability standards, but I say eff them. I resell my stuff to people who trust me, and they should get components that aren't abused. It keeps me in the upgrade game, so I don't where I eat

[Splave]

That doesn't sound lame at all. I've always, going back 10 years of ocing, have been a firm believer and advocate of real world stability. Running prime for 12 - 24 hours is just shaving off the top end of your systems lifespan all for the sake of proving something that you shouldn't have to. I've found, especially here, that some dudes get seriously bent out of shape when people refuse to meet their stability standards, but I say eff them. I resell my stuff to people who trust me, and they should get components that aren't abused. It keeps me in the upgrade game, so I don't where I eat

^this

[dejanh]

I must respectfully disagree; I really appreciate the work you've put into this thread and it's enlightened me quite a bit, but I beg to differ on this one. I think it's important on boards like this to have a certain set of standards regarding when an overclock is considered relatively stable for the purposes of spreading correct information regarding expectations. While I completely agree with your assessment that 24/7 for a month doesn't necessarily equate to stability, I also believe it's important to use the most scientific means at our disposal to render judgment on what may or may not be considered stable. I also agree that having alot of experience overclocking can lead to having a good "gut feel" of stability, but one must remember that alot of folks in this forum (albeit not as much as other forums) are newer and looking for advice. Developing that "gut feel" takes years of practice.

That being said, I would like to recommend to anyone seeking advice that anything in the area of 12 hrs of prime or 6 hrs of LinX goes a long way to "prove" stability.

Comments welcome, but I don't want to hijack this thread; so try not to pay too much attention to me.

On that note, I just got my P6T deluxe/920/6GB OCZ 1600 Plat. Seems to be a great chip, I could hit 3.6Ghz on water at just above stock volts. I'm still learning this platform and will post results when I get anything substantial.

Thanks!

I don't think you are disagreeing with me, and I did not say that we should not have a certain set of standards that determine a stable OC, especially for those that are new to the game. You have to start somewhere and with something. However, over time you evolve an inherent knowledge of "what works and what does not", and I will take that any day over any Prime/Linpack test. The best test for stability is one where you take real world applications and you task them to do the work on the CPU and see if is stable. For example, if you use your machine to fold for 2/3 of the day or more, then use F@H to test your stability, not Prime.

If you look carefully in this thread, there are people that get all bent out of shape here that after 10 hours of running Prime it fails, or worse yet, after like 20 hours it fails. No need to get bent out of shape. At some point in time Prime, Linpack, all stress tests will cause your system to fail. This would be considered "reasonable failure". I heard a good comment some 10 years ago from some (at the time) prominent overclocker that said "run Prime for 20 minutes to 1 hour and you'll likely never see your system fail". I agree that with today's multicore chips things are bound to be a bit different but in essence this has not failed me since, especially when I coupled it with real world tests.

Anyway, to stop the thread from derailing more, let's just get back on topic

[highoctane]

Anyway, to stop the thread from derailing more, let's just get back on topic

Well I wouldn't call it thread derailing, OC'ing i7/x58 and testing stability to some degree go hand and hand.

I'm sure we can all agree we do what we do to our own standards for our own peace of mind to ensure stability of the core system as well as the integrity of the data that system will be processing since poor stability can easily corrupt data as well as crash the OS, sometimes slowly sometimes quickly.

I've had 24hr "prime stable" rigs start crashing slowly after 72hrs or more of normal operation. As in strange things happening with the OS whether it be apps hanging/pausing randomly, not being able to simply disconnect a usb drive via safely remove, to simply loosing tcpip until a reboot but again these issues don't always develop in 24, 48 or 72 hrs or under prime loads.

So for those that more frequently reboot or turn their rigs on and off they may never run into these issues but for those running say a month at a time longer stability testing in every regard becomes more of an issue.

The level of stability testing is really a matter of choice like insurance, you can decide how much coverage you want/need but just because that policy works for some it won't neccisarily apply to others.

[dejanh]

Well I wouldn't call it thread derailing, OC'ing i7/x58 and testing stability to some degree go hand and hand.

I'm sure we can all agree we do what we do to our own standards for our own peace of mind to ensure stability of the core system as well as the integrity of the data that system will be processing since poor stability can easily corrupt data as well as crash the OS, sometimes slowly sometimes quickly.

I've had 24hr "prime stable" rigs start crashing slowly after 72hrs or more of normal operation. As in strange things happening with the OS whether it be apps hanging/pausing randomly, not being able to simply disconnect a usb drive via safely remove, to simply loosing tcpip until a reboot but again these issues don't always develop in 24, 48 or 72 hrs or under prime loads.

So for those that more frequently reboot or turn their rigs on and off they may never run into these issues but for those running say a month at a time longer stability testing in every regard becomes more of an issue.

The level of stability testing is really a matter of choice like insurance, you can decide how much coverage you want/need but just because that policy works for some it won't neccisarily apply to others.

Well, if you need months of uninterruped stability at full load 24/7 but shopping desktop parts, you are buying the wrong stuff and overclocking should most certainly not be your game.

This is mostly a business for hobbyists, enthusiasts, and gamers. You'll never see an overclocked server farm. That'd just be dumb.

The reason why I said not to derail the thread is because this can easily turn into an argument over nothing resulting in complete neglect of the original point of the thread. We already established what we all think, some of us agree while others can agree to disagree.

That's it.

[Splave]

On another note, was running prime had a nice 8 hours stable so far and GF turned the computer off when I went on lunch at work lol
As long as I know it was stable

4.3ghz turbo on no HT at 1.405v core CPU-z loaded

[ReverendMaynard]

On another note, was running prime had a nice 8 hours stable so far and GF turned the computer off when I went on lunch at work lol
As long as I know it was stable

4.3ghz turbo on no HT at 1.405v core CPU-z loaded

Boo-urns man. 4.3ghz stable is no joke HT on or off and 1.405v after vdroop is pretty impressive.

[dejanh]

How is this for a low voltage OC on air? 3.8GHz (19x200MHz), HT on, 1.2V, 1600MHz DRAM 7-8-7-20-1T, 1.35V QPI/DRAM, 1.65V DRAM ? ? Let's hear it...

[ReverendMaynard]

How is this for a low voltage OC on air? 3.8GHz (19x200MHz), HT on, 1.2V, 1600MHz DRAM 7-8-7-20-1T, 1.35V QPI/DRAM, 1.65V DRAM ? ? Let's hear it...

Looks fantastic dej. Is this a new chip, and how does it scale vcore/qpi wise up to 4.2ghz?

[dejanh]

Looks fantastic dej. Is this a new chip, and how does it scale vcore/qpi wise up to 4.2ghz?

Well, the scaling seems almost linear past 3.8GHz, at least for the next two multipliers (20x and 21x). I need about 0.1V - 0.12V on the core per every extra 200MHz (+1x multi.) using the same settings as above for the rest of the system (i.e., HT on, same DRAM settings). I can almost guarantee that it is my cooling having an effect on it as well. I know from the previous two i7 940s I tested that if cooled better lower voltages are possible. The difference is not huge, but still, it is maybe 1.38V for 4.2GHz if cooled better vs. 1.4V. A difference nevertheless. At the moment though it is how I said in the beginning...about 0.1V per additional 200MHz. No other voltages, including QPI change at all.

Oh and yes, this is my new i7 940.

[YukonTrooper]

How is this for a low voltage OC on air? 3.8GHz (19x200MHz), HT on, 1.2V, 1600MHz DRAM 7-8-7-20-1T, 1.35V QPI/DRAM, 1.65V DRAM ? ? Let's hear it...

I'll trade you chips when my 920 gets here. I won't even open it up to test.

[GenTarkin]

I think Ive seen the intel max voltage specification image for the i7 in this thread, but I dont know what page it was on...tried searching for all sorts of words to make it come up but cant find it. Was wondering would anyone be so kind as to post that image up for me again? I just want to see what max safe volts for all parts of this chip are. Before I start clocking her =)

On another note, Ive read your post dejanh about the relationship between qpi, uncore and ram speeds. I dont think it mentioned this anywhere in there, but is it possible to run qpi/uncore ABOVE core clock speed?
Like, say I get my chip to 3.3ghz or something mild....can I run qpi near below 4ghz but then run uncore at like 3.5ghz?

Thanks

[ReverendMaynard]

I think Ive seen the intel max voltage specification image for the i7 in this thread, but I dont know what page it was on...tried searching for all sorts of words to make it come up but cant find it. Was wondering would anyone be so kind as to post that image up for me again? I just want to see what max safe volts for all parts of this chip are. Before I start clocking her =)
Thanks

off the top of my head it was 1.55v vcore, 1.35v vtt and after that...it gets hazy.

[dejanh]

I think Ive seen the intel max voltage specification image for the i7 in this thread, but I dont know what page it was on...tried searching for all sorts of words to make it come up but cant find it. Was wondering would anyone be so kind as to post that image up for me again? I just want to see what max safe volts for all parts of this chip are. Before I start clocking her =)

On another note, Ive read your post dejanh about the relationship between qpi, uncore and ram speeds. I dont think it mentioned this anywhere in there, but is it possible to run qpi/uncore ABOVE core clock speed?
Like, say I get my chip to 3.3ghz or something mild....can I run qpi near below 4ghz but then run uncore at like 3.5ghz?

Thanks

Here is where you find all of the i7 datasheets. The information you are looking for will be in the Processor Datasheet - Volume 1.

http://www.intel.com/design/corei7ee/documentation.htm

Maybe this link should be permanently added to the first post.

In terms of running QPI/Uncore higher than the CPU clock, yeah, you should be able to do so. I do not see a reason why this would not be possible. Even by default the QPI runs higher than CPU clock and Uncore will run same or higher than the lowest clocked i7 920 as soon as you use 1333MHz DRAM because Uncore must be at least 2x DRAM.

[speckled10]

That doesn't sound lame at all. I've always, going back 10 years of ocing, have been a firm believer and advocate of real world stability. Running prime for 12 - 24 hours is just shaving off the top end of your systems lifespan all for the sake of proving something that you shouldn't have to. I've found, especially here, that some dudes get seriously bent out of shape when people refuse to meet their stability standards, but I say eff them. I resell my stuff to people who trust me, and they should get components that aren't abused. It keeps me in the upgrade game, so I don't where I eat

Amen to that Reverend !!

[Hornet331]

Anyone knows where the clock generator for the bclck on the mobo is located?

I thought about something regading the 222mhz bclck limit.

What if this limit is not due to the qpi clock but rather caused by the bclck generator itself?

This would also explain why all boards can clock near that limit, some higher, some lower.

Maybe a voltmod for the clock generator would harness higher clocks?

Or it also could be a complete brainfart from my side.

[dejanh]

Anyone knows where the clock generator for the bclck on the mobo is located?

I thought about something regading the 222mhz bclck limit.

What if this limit is not due to the qpi clock but rather caused by the bclck generator itself?

This would also explain why all boards can clock near that limit, some higher, some lower.

Maybe a voltmod for the clock generator would harness higher clocks?

Or it also could be a complete brainfart from my side.

Actually...that makes no sense.

If it was the clock generator of some sort causing the 222MHz BCLK limit then it would surely be different across different manufacturers. But, if it is a limitation of the CPU it would show up on all boards, from all manufacturers, regardless of the board engineering and would have very slight variations (few MHz up and down).

Remember this, and then maybe we can revisit it sometime...

If Intel unlocks new QPI multipliers, particularly lower than 18x (16x and 14x would be nice) and unlocks the 20x multiplier we will be able to push the BCLK much higher. I heard that this is in the works in the likes of a BIOS patch, but I do not have this officially confirmed.

[Hornet331]

Actually...that makes no sense.

If it was the clock generator of some sort causing the 222MHz BCLK limit then it would surely be different across different manufacturers. But, if it is a limitation of the CPU it would show up on all boards, from all manufacturers, regardless of the board engineering and would have very slight variations (few MHz up and down).

Remember this, and then maybe we can revisit it sometime...

If Intel unlocks new QPI multipliers, particularly lower than 18x (16x and 14x would be nice) and unlocks the 20x multiplier we will be able to push the BCLK much higher. I heard that this is in the works in the likes of a BIOS patch, but I do not have this officially confirmed.

Well im not exactly sure if QPI clocks are really the culprit here, cause of the slowmode some boards offer. QPI clock is a lot slower then with this option, yet still the cpu has the same QPI limit.

Also we dont know which clock generator everyone uses, but the chance is high that all board manufacturors use the same clock generator from the same company. Ci7 is new teritory and usually suppliers of components are limited at the beginning of a new product.

Anyway, its just a random idea. We'll see soon enough i guess.

[GenTarkin]

Here is where you find all of the i7 datasheets. The information you are looking for will be in the Processor Datasheet - Volume 1.

http://www.intel.com/design/corei7ee/documentation.htm

Maybe this link should be permanently added to the first post.

In terms of running QPI/Uncore higher than the CPU clock, yeah, you should be able to do so. I do not see a reason why this would not be possible. Even by default the QPI runs higher than CPU clock and Uncore will run same or higher than the lowest clocked i7 920 as soon as you use 1333MHz DRAM because Uncore must be at least 2x DRAM.

Thanks for that link, I appreciate it =)

[dejanh]

Well im not exactly sure if QPI clocks are really the culprit here, cause of the slowmode some boards offer. QPI clock is a lot slower then with this option, yet still the cpu has the same QPI limit.

Also we dont know which clock generator everyone uses, but the chance is high that all board manufacturors use the same clock generator from the same company. Ci7 is new teritory and usually suppliers of components are limited at the beginning of a new product.

Anyway, its just a random idea. We'll see soon enough i guess.

Well, I decided to inquire about this and according to one of my hardware engineer friends it is very likely that the clock generator is actually located on the CPU itself which would eliminate this as the possiblity.

[cstkl1]

Now with TRUE

[dejanh]

Here is an updated screenie 20-pass Linpack run...

And before somebody else asks me, power saving transition states are active so the CPU multiplier changes when not under load. However, while under load the CPU is running at full 3.8GHz

[SteveRo]

I think Ive seen the intel max voltage specification image for the i7 in this thread, but I dont know what page it was on...tried searching for all sorts of words to make it come up but cant find it. Was wondering would anyone be so kind as to post that image up for me again? I just want to see what max safe volts for all parts of this chip are. Before I start clocking her =)

On another note, Ive read your post dejanh about the relationship between qpi, uncore and ram speeds. I dont think it mentioned this anywhere in there, but is it possible to run qpi/uncore ABOVE core clock speed?
Like, say I get my chip to 3.3ghz or something mild....can I run qpi near below 4ghz but then run uncore at like 3.5ghz?

Thanks

some of the max v's here -

[Hornet331]

Well, I decided to inquire about this and according to one of my hardware engineer friends it is very likely that the clock generator is actually located on the CPU itself which would eliminate this as the possiblity.

I had the fear that the clock generator would be located on the cpu.