Overclocking (in)stability (Q6600)

[lefenzy]

I am trying to overclock a Q6600 Core 2 Quad (2.4 GHz) on a MSI nForce 650i Ultra board. I have found that the processor is stable at 333 FSB or 3 GHz using Prime95/LinX. Stock CPU voltage on the processor seems fine; perhaps even undervolting with RMClock is possible. I add 0.05V to the northbridge. Memory are 800 MHz total 6 GB; the value ram sticks have worked at 889 MHz too.

But the trouble is that the computer does not cold boot. Either it won't POST or it'll restart when loading windows. If I leave the computer on for a few minutes however, for example just by spending time investigating the BIOS settings, the computer can boot into windows successfully, and I can put it through stress testing and such. Now I haven't done testing as rigorous as 24 hr Prime95, but I do not know why cold booting does not work when I try. Here is a typical case: I shut the computer off for 20-30 minutes, and then I cannot boot successfully. Let the computer stay on, and it works.

I've tried increasing CPU voltage, NB voltage, memory voltage, disabling speedstep, but these do not appear to have worked.

Does anyone have any suggestions as to the cause of this kind of instability?

Could it be a power supply warmup issue? The power supply is a Corsair brand 430W. I think it should be fine.

 

[Schmide]

Is your memory set as close to 1 to 1 as possible?

 

[Phynaz]

Is it stable if you back off the overclock a bit? What happens at 2.9Ghz?

 

[lefenzy]

@ Schmide. The stock memory is not capable of reaching the high frequency of the FSB, so there is a divider. I.e. stock 400 MHz DDR2 makes for 5:6 or 444 MHz memory is 3:4 right?

@ Phynaz. I will have to do that. I just wanted an easy overclock to 3 GHz. Nice round clock frequencies. Perhaps I will try 2.7 and 300 FSB. But you're right I must diagnose my issues with experimentation.

 

[2is]

I've heard those nForce chipsets aren't the best for overclocking quads.

 

[Schmide]

The gist is this.

For a 266 processor like the q6600

Running the memory at

533 is 1:1
667 is 5:4
800 is 3:2

So if you raise the FSB from 266 to 333 the memory will be running at

667
833
1000

Respectively.

So if you raise the FSB from 266 to 400 the memory will be running at

800
1000
1200

 

[BonzaiDuck]

The gist is this.

For a 266 processor like the q6600

Running the memory at

533 is 1:1
667 is 5:4
800 is 3:2

So if you raise the FSB from 266 to 333 the memory will be running at

667
833
1000

Respectively.

So if you raise the FSB from 266 to 400 the memory will be running at

800
1000
1200

Groggy as I am this morning, I'll throw in on this because I went through the hoops in 2007 for a Kentsfield core. The OP didn't mention which stepping he has. Mine was the B3; in July '07 they released the G0 stepping, which had some advantages.

The Kentsfield, for a "practical" OC that was also "easy" to achieve -- could be guar-an-TEED a 3.0 Ghz over-clock.

The most practical way to achieve this was either to run the RAM at DDR2-667, or use ratios like 3:4 or 4:5. It may have been that the 4:5 ratio was the most efficient of the latter two -- I'm not sure now (my memory is stale).

With the B3 stepping, you indeed had to boost the VCORE a tad. It should at least be somewhere between 1.30V and 1.32V. With the G0 stepping, the voltage requirements were found to be less.

HOWEVER -- that leads to an observation I made on a particular nVIDIA chipset and BIOS. At a point before release of the G0 stepping, a BIOS version had been posted which (in hindsight) seemed ideal for the B3 stepping. Over a few months following release of the G0, additional BIOS revisions were released which seemed to be great for the G0 but which were terrible for the B3. (Or -- you needed to find new over-clock settings by trial and error instead of resetting to the old ones.)

I had always guessed that the voltage demands were highest at boot-up. And the OP also says he's using one rated at 430W. Since I don't know what else he has in the box, I'll let him reassess on his own.

ONE MORE POINT. If the motherboard is "used" and is not built entirely with solid-state capacitors and other components, there may be a capacitor-aging issue. I formed my own conclusions on this: In order to get a real-time "reported' load voltage of ~1.31V, you had to push the "set" voltage eventually to 1.375xx or even just a tad higher. this became apparent to me after I was running the OC'd 3.0 GHz configuration with a 680i chipset after a year or so. Since then, as far as I know, there has been little change with that issue. [It's my brother's computer now . . . ]

 

[john3850]

These mb only worked correctly with c2d and were nitemares with the quads.
I had one of the first evga 680i mb so I could run sli.
Evga even gave every one a free new updated mb 9 months later.
Evga then offered everyone new 780 mb for $60 bucks to support the later quads.

I never knew how bad and slow my evga mb was untill I got my first intel board.

 

[BonzaiDuck]

These mb only worked correctly with c2d and were nitemares with the quads.
I had one of the first evga 680i mb so I could run sli.
Evga even gave every one a free new updated mb 9 months later.
Evga then offered everyone new 780 mb for $60 bucks to support the later quads.

I never knew how bad and slow my evga mb was untill I got my first intel board.

The disparity in compatibility was between the later Wofldale and Yorkfield cores -- for instance, the 680i would work with the Wolfdale with a BIOS upgrade, but was totally hopeless or impossible with the Yorkie.

The earlier Conroe C2D and Kentsfield (C2Qs) worked fine with the 680i. I can't fathom why it wouldn't work fine with a 650i.

Also, now that you mention it, there was some problem with the nVidia chipsets for certain manufacturers earlier in the game -- I think eVGA was among them. But that was even before the milestone of the Wolfie and Yorkie releases. My ASUS 680i was excluded from the list of problem boards -- with specific assurances in various forums and web-sites.

 

[soccerballtux]

probably at least somehow related to the nForce. nForce absolutely sucked for overclocking Core2s.

 

[lefenzy]

Well I think I should be ok now. I never played with the FSB VTT setting. I have a MSI P6N SLI board, which confusingly has the VTT in increments of 2% between 0 and 20%. So although I do not know the numbers, after raising the setting to 2%, I am able to cold boot. Stress testing right now, but I think 3 GHz/333 MHz FSB/889 MHz memory will work for me. CPU VID is 1.2750 V, but I have lowered the voltage using RMClock to 1.25 V so cooler temperatures for me as well. I am indeed running a G0 stepping proc.

 

[soccerballtux]

ah ook great!
now time to go to 3.4ghz. Should be pretty simple and you'll be nearly future proof.

 

[Schmide]

(strike this) No G0 will hit 3.4 without exotic cooling.

 

[BonzaiDuck]

No G0 will hit 3.4 without exotic cooling.

I think I was able to get my B3 to 3.2+, where it stayed for a few months before I trimmed it back. I'd heard that the G0 made it possible to go higher -- maybe 3.4 -- and I think I saw examples of "air" (heatpipe) cooling. But that's not the main issue here.

If he wants it to run at 3.0, the lower voltages -- from what he says -- are shown.

Indeed, the VTT voltage on my B3 needed a bit of tuning, so if he "missed it" earlier, he may find his way to a stable OC at this point. I found it more expedient to run my memory to no more than ~ 850 (DDR). If he can do this, and it's stable, that's . . . . great . . . too . . .

 

[Schmide]

Man I feel dumb. The G0 was the good stepping. Sry. I had one, my nephew has it as an extra computer now, that runs at 3.5, I think I took it back to 3.0 to extend it's life.

 

[BonzaiDuck]

Man I feel dumb. The G0 was the good stepping. Sry. I had one, my nephew has it as an extra computer now, that runs at 3.5, I think I took it back to 3.0 to extend it's life.

Supposedly the official, spec maximum of the "safe voltage" range for that processor is around 1.36 to 1.37V, and obviously, if you needed to do it, 1.38 probably would not hurt it. One of us, speaking of the B3 stepping some years back, thought that 1.44V was a limit beyond which you should not venture -- with either WC or air-cooling -- either way.

But why run a G0 like that? For every voltage setting and speed of a B3, you can get the same speed and lower voltage with the G0 -- I think, anyway . . .

It's actually good to see people playing with these yesteryear chips and motherboards. I have a friend -- retired electronics guy -- who really pinches pennies. So he's playing around with three Dell C2D systems he bought on E-bay, but couldn't resist buying a Samsung 830 at a fire-sale price . . . .

 

[Idontcare]

Supposedly the official, spec maximum of the "safe voltage" range for that processor is around 1.36 to 1.37V, and obviously, if you needed to do it, 1.38 probably would not hurt it. One of us, speaking of the B3 stepping some years back, thought that 1.44V was a limit beyond which you should not venture -- with either WC or air-cooling -- either way.

Intel max voltage spec for Q6600 is 1.55V.

Download spec here, see Table 3 on page 17.

I ran my B3 stepping Q6600 right at 1.55V and 4GHz, albeit under vaporphase cooling, at full load 24x7 for nearly 2 yrs before it went kaputz from shorting out one too many times from the ice. (funny thing, Intel doesn't have a "no ice" spec for their processors )

 

[BonzaiDuck]

Intel max voltage spec for Q6600 is 1.55V.

Download spec here, see Table 3 on page 17.

I ran my B3 stepping Q6600 right at 1.55V and 4GHz, albeit under vaporphase cooling, at full load 24x7 for nearly 2 yrs before it went kaputz from shorting out one too many times from the ice. (funny thing, Intel doesn't have a "no ice" spec for their processors )

Well, I don't have the intel pages in front of me, but they used to cite a "safe range" and an "operable range" -- the first being a subset of the second. And I thought for sure the upper limit of "safe" was as I cited. Maybe that was for the Wolfie and Yorkie cores, though, but it was a fairly common number -- possibly spanning different gen chips . . . 1.3625V . . . etc. Are you SURE it was just the condensation/ice that caused the kaputz?

I think I remember reading that you could run a chip within "operable," and it would begin to show degradation. At which point, if you dropped the voltage back, it would continue to "work OK" -- but "not the same ever again."

The 1.44 number had come from AigoMorla, whom you would remember . . . I just haven't seen him posting here for some time. Maybe he decided that "enthusiast-computing" was an addiction, and he "went on the wagon" -- deciding to start a new phase of life. I dunno . . . Heh-heh! "A phase change" . . . .

 

[Idontcare]

Well, I don't have the intel pages in front of me...

Ah c'mon man I even went to all the trouble of tracking down the specific page and table, including providing a link to the Intel document spec itself, in my post

Intel max voltage spec for Q6600 is 1.55V.

Download spec here, see Table 3 on page 17.

That said, yes, Intel does this thing where they "revise upwards" their processor specs over time as they come to flesh out their datasets with more and more data points fed back from real-world usage scenarios. It would not surprise me if the "original" spec called for a max Vcc of 1.35V.

Regardless, as it stands now, Intel feels comfortable in asserting 1.55Vmax for Q6600 cpus. However, as you and I both know, the question one should ask is "why on earth would you want to shove 1.55V through your Q6600?".

At -40C, my Q6600 was gulping down ~270W at 1.55V...on air that probably would have been closer to 400W once you add in the temperature-induced leakage.

So 1.55V may be the max that Intel is comfortable covering in their 3yr warranty but that doesn't mean one can use that max Vcc in a practical sense because of the cooling requirements involved. (presumably the right kind of high-end water cooling setup could handle it though)

Also, as you mention there are "functional" limits and then "max" spec limits. The functional limit for Q6600 is 1.5V, the max spec is 1.55V, the difference is as you say, Intel states that operating in excess of the functional limit while still operating within the max limit will result in loss of long-term reliability (i.e. notable degradation can be expected within the warranty period).

Are you SURE it was just the condensation/ice that caused the kaputz?

I saw the sparks and arcing with my eyes, the ice wasn't the issue per se, defrosting it and forgetting to unplug the powersupply from the outlet was the problem.

A few of the socket pins were completely melted, and the corresponding sites on the CPU were blackened. The CPU wouldn't post after that, tried two different mobo's just to be sure.

With Ivy Bridge we are entering a realm in which max voltage specs probably aren't going to be an issue because the chip reaches its TJmax long before reaching Vcc_max in a practical sense. Yeah you can be dumb and just set the voltage to stupid levels and watch your chip throttle all day long while also degrading, but its not the same kind of issue that enthusiasts have traditionally faced when OC'ing chips by over-volting them.

Even with an H100, my practical voltage limit with my 3770k is 1.3V, any higher and my chip hits its thermal throttling limit.

In summary - be it a Q6600 or a 3770k - good luck actually keeping the chip below it's TJmax with air cooling while operating the Vcc anywhere near its upper max spec limit.

 

[BonzaiDuck]

. . . That said, yes, Intel does this thing where they "revise upwards" their processor specs over time as they come to flesh out their datasets with more and more data points fed back from real-world usage scenarios. It would not surprise me if the "original" spec called for a max Vcc of 1.35V.
. . .
Regardless, as it stands now, Intel feels comfortable in asserting 1.55Vmax for Q6600 cpus. However, as you and I both know, the question one should ask is "why on earth would you want to shove 1.55V through your Q6600?".

At -40C, my Q6600 was gulping down ~270W at 1.55V...on air that probably would have been closer to 400W once you add in the temperature-induced leakage.

So 1.55V may be the max that Intel is comfortable covering in their 3yr warranty but that doesn't mean one can use that max Vcc in a practical sense because of the cooling requirements involved. (presumably the right kind of high-end water cooling setup could handle it though)

Also, as you mention there are "functional" limits and then "max" spec limits. The functional limit for Q6600 is 1.5V, the max spec is 1.55V, the difference is as you say, Intel states that operating in excess of the functional limit while still operating within the max limit will result in loss of long-term reliability (i.e. notable degradation can be expected within the warranty period).


I saw the sparks and arcing with my eyes, the ice wasn't the issue per se, defrosting it and forgetting to unplug the powersupply from the outlet was the problem.
. . . .

With Ivy Bridge we are entering a realm in which max voltage specs probably aren't going to be an issue because the chip reaches its TJmax long before reaching Vcc_max in a practical sense. Yeah you can be dumb and just set the voltage to stupid levels and watch your chip throttle all day long while also degrading, but its not the same kind of issue that enthusiasts have traditionally faced when OC'ing chips by over-volting them.

Even with an H100, my practical voltage limit with my 3770k is 1.3V, any higher and my chip hits its thermal throttling limit.

In summary - be it a Q6600 or a 3770k - good luck actually keeping the chip below it's TJmax with air cooling while operating the Vcc anywhere near its upper max spec limit.

News to me! So Intel itself wasn't sure until later that the chip was thermally limited and not voltage-limited? The hits just keep on comin'!

I still can't understand how Ice was allowed to form. Those phase-change kits have rubber grommets to prevent that -- or they're supposed to prevent it. I know the die isn't sealed from the environment by the processor-cap. But you'd think there'd be a way of insuring that the grommets provided an effective seal -- eliminating any source of condensation around the CPU or the board . . .

I'd looked into the phase-change kits years ago (well, 2006/2007). I just decided that the "successful project" with extreme cooling wasn't worth the money to me.

On the original spec -- you cite 1.35V. I know I first saw it on Graysky's sticky-- 1.36XXV -- on OC'ing before the sticky was removed. Then, later, I'm sure I saw it on an Intel white-paper, summary or spec sheet, explaining the difference between safe and operable limits. But nobody would've known for sure if those number were too conservative then, because of the thermal limits you cite.

As in the Eastwood western -- "man's got ta know his limitations . . " Person's got to weigh the life-cycle of a CPU generation versus the amount of initial investment -- unless you do it for the sheer joy with a bountiful budget . . .

 

[lopri]

I've heard those nForce chipsets aren't the best for overclocking quads.

Correct. I advise the OP to save time and run the CPU at stock. I have had plenty of experience with nForce 680i and 780i. Strongly advise against mating them with a quad-core CPU, or more than 3 HDDs. Also advise some active cooling on both the north bridge and the south bridge.

It is extremely difficult to track down the issue on these boards because a seemingly stable configuration at one moment can go south out of the blue very quickly. I concluded it a cursed platform for a quad after so many hours wasted.

 

[BonzaiDuck]

Correct. I advise the OP to save time and run the CPU at stock. I have had plenty of experience with nForce 680i and 780i. Strongly advise against mating them with a quad-core CPU, or more than 3 HDDs. Also advise some active cooling on both the north bridge and the south bridge.

It is extremely difficult to track down the issue on these boards because a seemingly stable configuration at one moment can go south out of the blue very quickly. I concluded it a cursed platform for a quad after so many hours wasted.

Been a while, Lo'!

As much as that may have been true more than not, I had a 680i, Q6600 B3, and four hDDs. How-ever! I was running the HDDs off a 3Ware RAID card . . . . Running since 2007, and still running at my bro's house . . .

One thing for sure -- it was a power-hog. And some otherboards were better than others. . . . too . . .

 

[lopri]

@Duck: I kind of get lazy during the summer time.

Is that your 680i board the first revision (EVGA or ASUS)? I can't believe NV got away with that piece of junk without a mass recall or a class-action suit. (Big kudos to EVGA for its outstanding customer service during that time) Admittedly no one whose job depended on maintaining stable systems would have purchased them, but it had ruined many enthusiasts' data with its faulty memory controllers and disk controllers.

 

[Shmee]

780i is fine for OCing Q6600s, at least the EVGA FTW version I had was. It took some work, but I had it at 3.6, just the temps were too high with the cooling I had. My regular OC (and now my friend's) is about 3.2 GHz on the board with a Q6600 G0. Keep in mind with quads, MCP overvolting may/probably will be required.

 

[lopri]

@Shmee: How many memory sticks did you use when you oc'ed the Q6600 to 3.60 GHz?

@OP: It may be helpful if you use only 2 sticks of memory.