Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

[dma0991]

Mine does 4.4 with 1.224V load, so sure.

Now I hate you.

 

[Ferzerp]

Wanna double hate me? It never passes 66C in IBT (on high or max memory settings, 10 runs).

 

[dma0991]

Wanna double hate me? It never passes 66C in IBT (on high or max memory settings, 10 runs).

Okay.

 

[Ferzerp]

Do you think that voltages in the 1.23 to 1.30V range are unreasonable? IDC posted some tables in this thread that pairs over-clock speeds and voltages. I think 4.7 Ghz required something around 1.29+V.

It's more that 4.6 requires something like 1.36V for mine. My wall is at 4.4Ghz.

 

[dqniel]

It's more that 4.6 requires something like 1.36V for mine. My wall is at 4.4Ghz.

Under Prime load:

4.5Ghz - 1.216v
4.6Ghz - 1.264v
4.7Ghz - 1.312v

SuperPi runs:

4.8Ghz - 1.312v
4.9Ghz - 1.352v

*updated*

 

[Ferzerp]

Cooling isn't an issue. I've run IBT at over 1.45V just to see. Sure it's in the 80's, but it's almost 1.5V!

 

[BonzaiDuck]

Cooling isn't an issue. I've run IBT at over 1.45V just to see. Sure it's in the 80's, but it's almost 1.5V!

That sounds strange. Dqniel's results are totally consistent with IDC's.

Others may know . . . Did we have recent exchanges here about "PLL Overvoltage?"

See, in the decades I've been building PC's, I don't think I ever got an "underperforming" Intel CPU. So . . . I'll leave you to others . . .

 

[Ferzerp]

I've had a smattering of both.

My i7-930 did 4.0Ghz at well under stock voltage (but wouldn't go higher).

 

[WhoBeDaPlaya]

I've had a smattering of both.
My i7-930 did 4.0Ghz at well under stock voltage (but wouldn't go higher).

Same here! My i7 920 D0 actually runs at a lower Vcore for it's max overclock compared to the 3570K
And it's not like its a single golden chip. My brother's 920 runs at exactly the same settings (same EX58-UD3R, Xigmatek DK, IC Diamond, both CPUs purchased at the same time from MC).

 

[ehume]

dqniel - Use a hacksaw or a mini-hacksaw and a pair of pliers to cut out your closed corners. Cut down each corner dam along each flange. Eight cuts leaves you with four tabs where the corner dams were. Then with a pair of pliers, reach the trough at the bottom of each tab. Grasping each tab with your pliers, bend. The base of the tab breaks off and you have open corners. That will allow you to mount your fan.

A link to a pictorial how-to is here.

 

[dqniel]

dqniel - Use a hacksaw or a mini-hacksaw and a pair of pliers to cut out your closed corners. Cut down each corner dam along each flange. Eight cuts leaves you with four tabs where the corner dams were. Then with a pair of pliers, reach the trough at the bottom of each tab. Grasping each tab with your pliers, bend. The base of the tab breaks off and you have open corners. That will allow you to mount your fan.

A link to a pictorial how-to is here.

Thanks for the link, but I've already done this before with a Dremel It's not worth the effort to me when additional Yate Loons with open corners are only $3.60 each. It's like an excuse to get more fans :whiste:

 

[ehume]

Thanks for the link, but I've already done this before with a Dremel It's not worth the effort to me when additional Yate Loons with open corners are only $3.60 each. It's like an excuse to get more fans :whiste:

By the time you had your Dremel out and plugged it in I'd be done cutting out the corners of a fan. And you can cut too much with a power tool. I take your point, though. I have more than a hundred fans.

 

[Rubycon]

I have more than a hundred fans.

That's fairly popular.

 

[dqniel]

I have more than a hundred fans.

I love your sound pressure vs temp testing in your sig. Very useful info, and makes me happy that I went with the YL H push/pull config. Seems like I probably got the best bang for the buck... decent performance for the noise they emit.

 

[mrob27]

Thanks for the link, but I've already done this before with a Dremel It's not worth the effort to me when additional Yate Loons with open corners are only $3.60 each. It's like an excuse to get more fans

By the time you had your Dremel out and plugged it in I'd be done cutting out the corners of a fan. And you can cut too much with a power tool. I take your point, though. I have more than a hundred fans.

And as everyone knows, the more fans in your rig, the quieter it'll run... :awe:

 

[Haserath]

And as everyone knows, the more fans in your rig, the quieter it'll run... :awe:

Only with Steve's magic.

What powerful magic it was.

 

[WhoBeDaPlaya]

I don't even have to mess with that, but I could if I wanted too (the options are there).

For me, I just leave LLC set to "Auto" in the BIOS. It is magically delicious.

I'm measuring Vcc with a voltmeter too, and it is rock-solid at whatever I set it at in AI Suite.

Apparently, setting LLC to 100% on the Asrock Pro4 doesn't mean what I thought it does.
100% - wild swings between Vcore idle and load (1.248V - 1.32V)
0% - MUCH more steady (1.248V - 1.262V)

 

[dqniel]

Apparently, setting LLC to 100% on the Asrock Pro4 doesn't mean what I thought it does.
100% - wild swings between Vcore idle and load (1.248V - 1.32V)
0% - MUCH more steady (1.248V - 1.262V)

Same with on the Pro4-M. For me, 50% yields minimal droop and pretty stable load voltage. 100% swings around a lot and droops further. I haven't tried 0%, but will now that I've read this.

 

[BonzaiDuck]

Apparently, setting LLC to 100% on the Asrock Pro4 doesn't mean what I thought it does.
100% - wild swings between Vcore idle and load (1.248V - 1.32V)
0% - MUCH more steady (1.248V - 1.262V)

Resurrected and recently much-mentioned is a fall, 2007 Anandtech article on overclocking the QX9650, which explains what happens with deployment of LLC.

While the article seems mildly reticent about using LLC or advises people not to use it, the main thing to take away from it is that it pushes upward a transient which exists in any case, cannot easily be seen or measured, and briefly increases the voltage to a level above the "maximum VID" which you set in BIOS. How those transients affect the life of your CPU is a matter of speculation, but become a greater concern the more you push the VCORE (load minimum, idle maximum) higher.

We've also discussed and concurred about the size of the transient spike under lighter-than-stress-test loading, comparing the interaction of LLC, VCORE and vDroop to a rubber-band. Under light load, the transients are correspondingly smaller and of less consequence.

That being said, the prevailing wisdom cautions against increasing LLC to a point beyond that which just exactly cancels out vDroop.

My own LLC setting eliminates about 50mV of vDroop out of a total 80mV, so load VCORE still droops by about 30mV below the idle "turbo" VCORE. If LLC helps to reduce maximum IDLE VCORE, then one can imagine that voltage spikes may go no higher than before you applied LLC.

 

[mrob27]

Resurrected and recently much-mentioned is a fall, 2007 Anandtech article on overclocking the QX9650, which explains what happens with deployment of LLC.

While the article seems mildly reticent about using LLC or advises people not to use it, the main thing to take away from it is that it pushes upward a transient which exists in any case, cannot easily be seen or measured, and briefly increases the voltage to a level above the "maximum VID" which you set in BIOS. How those transients affect the life of your CPU is a matter of speculation, but become a greater concern the more you push the VCORE (load minimum, idle maximum) higher.

[...]

I dug around and found the actual link:

Intel Processor Power Delivery Guidelines (Cont'd)

Personally, I think Intel's engineers are a lot smarter than most of us and that we should keep the peak voltage below VID.

 

[BonzaiDuck]

I dug around and found the actual link:

Intel Processor Power Delivery Guidelines (Cont'd)

Personally, I think Intel's engineers are a lot smarter than most of us and that we should keep the peak voltage below VID.

I think the point -- and please feel free to differ -- is that (1) the chip's "maximum VID" spec is above any "safe limit." (2) You set a maximum effective VID in BIOS when you tweak your voltages -- which should be well below the spec 1.5+V. (3) You're more at risk for transient "damage" the higher both VID and VCORE are set to rise.

Finally, the VRM and other features of the last two or so motherboard generations may (MAY) mitigate the risk further. But only the Intel engineers would know more specifically; instead, they say "don't let VCORE go higher than 1.XX V" [which is a rumor of what they may have said], or they publish a spec [which they don't for SB and IB while offering some sort of "tuning insurance" for $100].

On my Sandy Bridge, if I had idle VCORE maximum of 1.38V and load VCORE of 1.31+V before LLC, I might reduce the idle to 1.35V, set LLC to a moderate level, and still get the same peak voltages (including "that which cannot be easily measured.") In the process of doing this, I would've reduced VID to give me that 1.35V value. [Put another way, my maximum "set" VID in this scenario amounted to about 1.385V as shown in CoreTemp.]

In other words, load VCORE is slightly higher with less vDroop to give a stable value; idle VCORE has been reduced; the unmeasurable transient stays about the same. It would seem the sustained values are more relevant the lower they (and the transient) appear as compared to the max-VID spec (of 1.5+V). But the sustained "idle" will still be much lower with EIST; the sustained "loads" will be closer to the "turbo" idle under normal/light loading conditions; and the transient "bounce" under normal/light loading situations will be less.

You're going to be skirting around the edge of "unknown" territory anyway, when over-clocking. Chances are, if you pay attention to all these things and set modest objectives, your chip could last much longer than the warranty 3 years.

Of course, some "suggested safe limit" on the Ivy will be lower than the Sandy.

 

[dqniel]

At 0% LLC on my Pro4-M the vcore rises when under load. I don't like that. At 50% it makes it so the vcore is above .020v below the set vcore when under load. That seems like the safer yet still acceptable vdroop, so I'll leave the setting at 50%.

 

[BonzaiDuck]

At 0% LLC on my Pro4-M the vcore rises when under load. I don't like that. At 50% it makes it so the vcore is above .020v below the set vcore when under load. That seems like the safer yet still acceptable vdroop, so I'll leave the setting at 50%.

You mean "At 100% LLC . . . . vcore rises under load," don't you? I was very cautious about this with the SB when I started, preferring "No LLC," and then "25%" LLC. As long as it allows me to do what I did -- at same time reducing idle ("turbo") VCORE from 1.38 to 1.35, I feel pretty good about "50%."

Anyway, we diverge and digress. "TIM and de-lidding . . . "

 

[WhoBeDaPlaya]

Actually, on my Pro4 with 100% LLC, Vcore drops like a stone (~1.32V idle to ~1.24V) under load so I do not know WTF is going on (everything else, eg. Turboboost, Speedstep, C1E, C3, C6, etc is turned off).

I have to set LLC to 0% to get a more stable Vcore (~1.25V idle to 1.232V load). Have not tried 50% yet.

Anyhoo IDC, sorry about this :

 

[Ferzerp]

Resurrected and recently much-mentioned is a fall, 2007 Anandtech article on overclocking the QX9650, which explains what happens with deployment of LLC.

While the article seems mildly reticent about using LLC or advises people not to use it, the main thing to take away from it is that it pushes upward a transient which exists in any case, cannot easily be seen or measured, and briefly increases the voltage to a level above the "maximum VID" which you set in BIOS. How those transients affect the life of your CPU is a matter of speculation, but become a greater concern the more you push the VCORE (load minimum, idle maximum) higher.

We've also discussed and concurred about the size of the transient spike under lighter-than-stress-test loading, comparing the interaction of LLC, VCORE and vDroop to a rubber-band. Under light load, the transients are correspondingly smaller and of less consequence.

That being said, the prevailing wisdom cautions against increasing LLC to a point beyond that which just exactly cancels out vDroop.

My own LLC setting eliminates about 50mV of vDroop out of a total 80mV, so load VCORE still droops by about 30mV below the idle "turbo" VCORE. If LLC helps to reduce maximum IDLE VCORE, then one can imagine that voltage spikes may go no higher than before you applied LLC.

VID is never a value that you set.

VID is the value of voltage that the processor requests at a given speed. You cannot alter this.