vtt killed qx9650 and other 45nm cpus

[railmeat]

qx9650 dead,to much vtt voltage.$1000 gone..a really GOOD read
http://www.anandtech.com/weblog/showpost.aspx?i=428

For those that are requesting proof that excessive VTT was the cause of the failure, just know that our conclusions are based more on empirical evidence than anything else; however, we have had more than one conversation with Intel CPU power engineers regarding the practical limit of VTT. In general, Intel does an excellent job of providing a wide selection of both technical and educational documents available for download by anyone directly from their website. Oftentimes significant technical insight can be gained by merely taking a few moments to read through some of these publications. As an example, anyone wishing to know more about the 45nm voltage constraints as recommend by Intel can find more information here – http://download.intel.com/design/pro...hts/318726.pdf

Intel® Core™2 Extreme Processor QX9000 Series and Intel® Core™2 Quad Processor Q9000 Series Datasheet (Document Number: 318726-003) was first published in March 2008. Section 2, Electrical Specifications, Subsection 2.6, Voltage and Current Specifications, Table 2-2, titled Absolute Minimum and Maximum Ratings, specifies absolute maximum and minimum ratings for this series of processors. It’s important to note that these values lie outside the functional limits of these processors (i.e. they may not be indicative of warranted operating voltages but are rather intended to serve as warning to those that would run their processors above rated specifications).

Table 2-2 shows maximum voltages for both VTT (FSB Termination Voltage) and VCC (Vcore) with respect to VSS (the processor power ground plane voltage, a.k.a. “ground”). In this case they happen to be equal in magnitude – 1.45V. If anything, this suggests that the use of high VTT voltages can be potentially as damaging, if not more so, than the sustained application of Vcore voltages above this same voltage yet many users continue to operate under the assumption that excessive VTT voltages should be less concerning than higher Vcore voltages.

The attitude that only Vcore can damage a CPU is one that has prevailed for far too long. Many users where initially “shocked” to learn that high PLL voltages (used to provide power to the complex array of circuits used to distribute the CPU’s common clocking signal) could permanently damage or cripple a CPU with little or no advanced warning. Even today there are those that outwardly deny the possibility of failure due to the use of VTT well in excess of Intel’s published maximum specification.

We’ve been trying for a while now to advocate a more intelligent approach to overclocking. Our loss should serve as an example for all as to the potential consequences of ignoring a little good advice. If nothing else, please don’t let our mistake become your mistake, too.

guys on evga forum with 750i and 45nm leary of vtt voltages
http://www.evga.com/forums/tm.asp?m=...1&key=&#343283


Boot up speed comparison: 680i vs 790i
http://www.evga.com/forums/tm.asp?m=320538

[Leeghoofd]

Well I think Intel should have warned us beforehand... K to sum it up : max Vcore 1.45, CPU VTT 1.45, PLL voltage maybe max 1.7...

I still wonder how they determined the cause of death...

[railmeat]

Well I think Intel should have warned us beforehand... K to sum it up : max Vcore 1.45, CPU VTT 1.45, PLL voltage maybe max 1.7...

I still wonder how they determined the cause of death...

"I still wonder how they determined the cause of death"

ya how do u do a autopsy on a dead pc 0_o .

[cadaveca]

With the right tools, it can be done. How do you think they test early silicon?

[Charles Wirth]

Repost

Even with a good sink, 1.45v is a lot for 45nm for long term survival.

[Leeghoofd]

Even under water sir Fugger ? thing is temps really don't seem an issue with my Qx..; idling around 30°C and topping out at 55°C under 100% folding load... (if the readouts are correct ofcourse) at 4ghz with 1.25volts... the E8400 however needs allmost 1.4volts to be primestable at 4ghz... temps are way higher with that CPU almost reaching 65° but gonna back down till I put that one under water too...

[jonny_ftm]

This confirms what was already suspected. These chips die quickely on high vcore, at least quicker than 65nm chips. Again, loadline calibration and similar tools to bypass vdroop might give you a confortable feeling of security setting vcore to 1.4v, while, in real world, it hits 1.5v 1000x/sec without you know (overshoot)
Intel should have be more clear, but hey, no one would buy a 45nm if we can't OC to death :-)

[Leeghoofd]

Again, loadline calibration and similar tools to bypass vdroop might give you a confortable feeling of security setting vcore to 1.4v, while, in real world, it hits 1.5v 1000x/sec without you know (overshoot)

Is that on all mobo's , does that happen with all PSU's ? Dunno about that statement of yours is truly 100% certified...

I yet have to see a spike with OCCT during hours of abuse... the P5K Dlx remains pretty flat, only with the QX9770 I saw some very small variations, Striker Extreme ripples more though no weird erratic spikes observed yet... I know OCCT isn't bulletprrof proof either but I saw some with very bad spikes... till now pretty happy camper...

[jonny_ftm]

Oh, overshooting has nothing to do with PSU. Overvolting can be related to PSU though. Overshooting is the normal oscillations in voltage that occur thouthands time a second. They depend on the quality of the resistors and transistors on the motherboards essentially. Implementing a vdroop as a specification, gives motherboard manufactures a margin to use affordable material to keep the motherboard price into a reasonable range. When intel fixes 1.36v as a maximum for its 45nm CPUs, when bios autosets 1.36v on a CPU with that VID, a voffset occurs on idle and a vdroop on load.

Now, see it as the vcore you monitor with onboard sensors or a DMM is just a medium of thouthands/sec of microoscillations that you can't see without expensive oscilloscopes. Besides those micro-oscillations, on every load change, there will be spikes in the voltage delivered to the CPU that can't be sensed by a DMM, but only an oscilloscope. The most harmfull in overclocking is the ascending spikes (overshooting).
Implementing the vdroop and voffset gives the following:
For 1.36VID CPUs, on load, the motherboard vdroops (intentionally and by design) to 1.27v avearge, thus, when the CPU returns to idle from full load, an overshooting spike occurs but won't exceed intel specs of 1.36v.

Now, if you remove the security vdroop, when the CPU returns from load to idle, the overshooting will be 1.36+0.09v = 1.45v for quadcores

Removing vdroop by vmods alters further more the resistance and the overshooting spikes could easily exceed 0.1v.

The real solution is using very high quality PWM to deliver a stable current to the CPU even when sudden energy needs occur (load/idle frequent changing status)

Simpler way to know how good is your MB and how you're overshooting your CPU, try to OC without loadline calibration/vmod. When you get your stable OC, the bios set vcore will be what you at least reach when you enable loadline calibration and lower vcore.

Exp: My Q6600 needs 1.43v with loadline calibration to be stable on load at 3.8GHz. Without loadline calibration, I must set my bios to 1.53v to account for a 0.09 vdroop. So, by disabling vdroop, I'm exceeding intel specs of 1.50 max VID when my CPU goes from load to idle. In my case, I have a 65nm CPU and I'm exceeding intel specs by only 0.03v (only 2&#37

For someone using a Q9550 and setting 1.36v with loadline calibration enables, he overshoots up to 1.46v, that's 7% off intel specs. Above 10% specs, you're for sure killing your CPU slowly, make the math. Even on 65nm CPU, above 10% VID range gives 1.65vcore, and you know all that 65nm chips on 1.65vcore will die soon if not put on LN2/subzero temps

A good summerize is also here, but complicated, I tried to illustrate it
http://www.anandtech.com/cpuchipsets...spx?i=3184&p=5

[Leeghoofd]

Well too bad Asus boards don't lower voltages when using EIST... total CPU speed is lowered though Vcore remains the same... I would hate to have 1.4vcore going through the CPU at idle dropping to 1.35 levels or such when being under load... that doesn't seem good neither... overclocking is getting pretty complicated lol... and my PC is almost on 24/7...

[Gen]

I would hate to have 1.4vcore going through the CPU at idle dropping to 1.35 levels or such when being under load... that doesn't seem good neither... overclocking is getting pretty complicated lol... and my PC is almost on 24/7...

actually as jonny said it is a needed 'security' feature. for a thorough read on vdroop try this writeup made by freecableguy.

i no longer use load line calibration with my P5K-E.

[jonny_ftm]

actually as jonny said it is a needed 'security' feature. for a thorough read on vdroop try this writeup made by freecableguy.

i no longer use load line calibration with my P5K-E.

I personally still use loadline calibration to have a lower vcore at idle. Despite the Anandatec article showing more power consumption with loadline enabled, on Q6600 chips idle temperature is much more important if you disable loadline calibration. In my case, without loadline calibration, I have to set 1.53 bios, which gives me a voffset of 0.04v at idle (so 1.49vcore idle) and 0.09 vdroop for 1.43vcore load. On a quadcore, the CPU is most of the time idle depending on the PC use, so not using loadline calibration while overclocking is an energy loss, proven by the more important heat generated at idle.

Now, for moderate overclocks, I won't use load line calibration too

The most important is that people should account for the vdroop as being replaced by overshooting if they enable loadline calibration so that they don't burn their 45nm babies. On 65nm chips, before you exceed VID specs by 5-10%, you'll be limited by the cooling issues

[jg0001]

What is the default on VTT (?same as FSB volts?)

My mobo (Giga X48) only says +/- volts... If I set it to +0.2, what does that mean in real volts?

[Leeghoofd]

What is the default on VTT (?same as FSB volts?)

My mobo (Giga X48) only says +/- volts... If I set it to +0.2, what does that mean in real volts?

1.4 then I think 1.2 is stock for 45nm, but that doesn't necesarrily mean the board gives 1.4VTT could be more or even less...( sorry to confuse you even more) eg my Maximus SE undervolted CPU VTT... so if I set 1.3 it would have been around 1.23 level, so to get 1.3 in readout I needed about 1.4 set... your board could be spot on or too low or too high...

Isn't their hardware monitor or does it just say Ok in the voltage section of the Giga bios...

[railmeat]

1.4 volts should be good for water getting GREAT clocks,certain death looms at 1.45v + as fugger agrees.